---
title: 'Right-Side-Out: Asymmetry and Applications'
url: https://www.emergentmind.com/topics/right-side-out
type: topic
---

# Right-Side-Out: Asymmetry and Applications

“Right-Side-Out” designates several distinct research notions centered on orientation, asymmetry, or the methodological privileging of a “right” structure. In the cited literature, it appears literally as an exterior configuration to be identified or restored, as in Stegosaurus chirality and garment reversal; behaviorally as a preferred passing side in pedestrian motion; and analytically as a reformulation around a right-hand side, right action, or right-oriented geometric object in numerical linear algebra, PDE, optimization, coding theory, geometry, and gravitation [1611.08760], [1912.10870], [1601.03184], [1811.12908], [2208.13941], [2112.07322], [2509.15953]. This suggests a recurrent motif rather than a single discipline-specific definition: the term marks cases in which a privileged orientation or right-hand-side structure is operationally decisive.

## 1. Lexical scope and recurring technical pattern

Across these works, “Right-Side-Out” has three main meanings. First, it can denote a literal exterior state. In paleontology, it refers to an operationally defined chirality class of Stegosaurus, distinguished by the tilt of the largest dorsal plate. In robotics, it denotes the successful reversal of an inside-out garment into a right-side-out configuration [1611.08760], [2509.15953].

Second, it can denote a preferred side in behavior. In pedestrian dynamics, the relevant phenomenon is a statistically dominant tendency to walk on the right side in a symmetric conflict geometry, together with an explicit simulation parameter controlling side bias [1912.10870].

Third, it can denote a right-hand-side-centered reformulation. In DD-\(\alpha\)AMG, the setup is reorganized around multiple right-hand sides processed simultaneously; in elliptic and parabolic boundary Harnack theory, the novelty is to admit nonzero right-hand sides; in SOCP sensitivity analysis, the object of study is convexity in the right-hand-side parameter \(b\); in Gabidulin decoding, the algorithm acts on codewords on the right; and in gravitation, the emphasis falls on the matter source \(T_{\mu\nu}\) on the right side of Einstein’s equation [1601.03184], [1811.12908], [2307.06487], [2208.13941], [2112.07322], [1401.5513].

## 2. Exterior chirality and physical reversal

In “Stegosaurus chirality” [1611.08760], Right-Side-Out is an operational chirality class for an animal whose exterior arrangement is not superimposable on its mirror image. The paper defines exterior chirality by the staggered rows of dermal plates and by the fact that many individual plates are themselves chiral. The diagnostic criterion is the largest dorsal plate, identified as the 14th plate in the series: if that plate tilts to the right, the specimen is designated \((R)\); if it tilts to the left, it is \((L)\). Using plate-surface identity, plate tilt relative to the midline, and asymmetry of plate bases, the authors assign USNM 4394, USNM 4714, DMNS 2818, and NHMUK R36730 to the \((R)\) form. They also identify several mirror-image confusions in reconstructions and mounts, including swapped left and right plate surfaces and displays implying anatomically impossible mirrored individuals. On that basis, the paper argues that museum mounting and illustration should explicitly record chirality, and it regards the predominance of the \((R)\) form as consistent with the hypothesis that the plates acted primarily as display structures.

In “Right-Side-Out: Learning Zero-Shot Sim-to-Real Garment Reversal” [2509.15953], the phrase returns to a literal inversion task in robotics. The problem is defined for inside-out upper garments on a table and is difficult because of hybrid quasi-static and dynamic phases, rapid contact changes, self-collision and friction, and severe visual occlusion. The framework decomposes the task into Drag/Fling, which creates and stabilizes an access opening, and Insert&Pull, which inverts the garment. Each primitive is keypoint-parameterized from depth and executed by fixed bimanual motion templates. The perception stack uses U-Nets on overhead depth and a binary garment mask, with D3RoMa used during deployment to refine depth. Training data are generated in a GPU-parallel Material Point Method simulator with thin-shell deformation, codimensional friction, and batched rollouts, and no human annotations are used. Success is defined by right-side-out coverage from a single overhead frame, with an episode counted successful if coverage is at least \(0.80\). Policies trained entirely in simulation deploy zero-shot on real hardware with a single depth camera and achieve up to \(81.3\%\) success. The two studies are technically unrelated, but both treat right-side-out not as a metaphor but as an externally verifiable configuration.

## 3. Side preference as a behavioral convention

In pedestrian dynamics, the relevant usage concerns right-side preference in motion rather than object orientation. “Go left or right? Explore the side preference behavior with circle antipode experiments” [1912.10870] studies pedestrians walking simultaneously from points on a circle to their antipodes. The experiments were conducted outdoors at Beijing Jiaotong University in December 2017 with two circle radii, \(r=5\) m and \(r=10\) m, and with \(N=8,16,32,64\) participants per trial, each experimental type repeated four times for 32 trials in total. From 960 trajectory samples, \(68.75\%\) were classified as right-side preferred and \(31.25\%\) as left-side preferred.

The paper reports that handedness could not be robustly tested because only three participants were left-handed, while gender and height showed no significant effects: the Mann–Whitney \(U\) test for gender gave \(p=0.157\), and the Kruskal–Wallis \(H\) test for height groups gave \(p=0.496\). The timing analysis indicates that side choice is made very early. At departure distance \(d=1\) m, consistency between local and global side classification already exceeded chance and was above \(70\%\) in the 10 m circles and about \(77.5\%\) in the 5 m circles; consistency rose to about \(90\%\) at roughly one-third of the circle radius. This is interpreted as evidence that side choice is part of an early route strategy rather than a late correction.

The same study also links side preference to efficiency and simulation. Travel time \(T\) was the sole efficiency metric, and right-side choice produced significantly shorter travel times in several scenarios, including 5m-8p, 5m-32p, 5m-64p, and 10m-8p. To reproduce the observed bias, the authors modify a Voronoi-diagram-based navigation model by introducing a side-preference exponent \(\gamma\), calibrated as \(\gamma \sim N(0.25,0.5^2)\). With this distribution, simulated right-side fractions fell in \(0.664\)–\(0.766\), and 53 out of 64 simulated individuals showed right-side preference. Here, right-sidedness is neither morphological nor algebraic; it is a population-level convention embedded in route selection.

## 4. Multiple right-hand sides in high-performance computation

In numerical linear algebra, Right-Side-Out denotes a reordering of work around multiple right-hand sides. “Multiple right-hand-side setup for the DD-\(\alpha\)AMG” [1601.03184] concerns the setup phase of DD-\(\alpha\)AMG, a two-level adaptive algebraic multigrid preconditioner for the lattice QCD Dirac equation. The central modification is to process many right-hand sides simultaneously instead of one after another. In the setup’s iterative refinement, test vectors are therefore updated blockwise rather than sequentially, so that operations such as restriction, coarse solve, prolongation, and smoothing become block operations.

This change alters the computational regime. Sparse matrix-vector operations are replaced by sparse matrix-matrix operations, raising arithmetic intensity and improving cache reuse. Halo exchanges and global sums are aggregated over a block of size \(b\), increasing message size by a factor \(b\) and moving communication away from a latency-dominated regime. The implementation described in the paper uses an Array-of-Structs-of-Short-Vectors layout keyed by right-hand side, with \(b\) chosen equal to the SIMD width, specifically \(b=16\) on Intel Xeon Phi KNC. On the reported hardware and lattice, the measured speedups in setup components were about \(2.9\times\) for restriction/prolongation, \(2.4\times\) for coarse-grid computation, \(4.7\times\) for halo exchanges, and \(10.3\times\) for global sums. The coarse-grid portion of the setup was reduced by about \(2.9\times\), and the end-to-end setup was improved by about \(1.4\times\) in the partially converted implementation; the abstract states that the parts implemented so far show a speedup of roughly \(3\times\) compared to the optimized single-right-hand-side setup.

The mathematical content is unchanged: the same near-null test vectors define the same \(R\), \(P=R^\dagger\), and coarse operator \(A_c=RAP\). The improvement is architectural and algorithmic rather than variational. In this usage, Right-Side-Out means turning the setup “inside out” so that right-hand-side multiplicity becomes the organizing principle of the computation.

## 5. Right-hand-side reformulations in PDE and optimization

In elliptic boundary regularity, the novelty of “A new boundary Harnack principle (equations with right hand side)” [1811.12908] is precisely the inclusion of forcing and lower-order terms in a boundary Harnack principle. The operator is
\[
L u := \operatorname{div}(A(x)\nabla u) + b(x)\cdot \nabla u + c(x)u,
\]
with bounded measurable real-symmetric \(A\), uniform ellipticity, bounded \(b\) and \(c\), and \(c\le 0\). The paper proves boundary Harnack comparability in Lipschitz domains for functions with \(|Lu(x)| \le d(x)^\gamma\), where \(d\) is distance to the relevant boundary portion, provided the compatibility condition \(2-\alpha+\gamma>0\) holds. In cones, the analogous threshold is \(2-\alpha_1+\gamma>0\), where \(\alpha_1\) is the homogeneity of the positive harmonic function vanishing on the boundary. The argument proceeds through barriers, blow-ups, compactness, and Liouville-type results. The inclusion of a zero-order term is emphasized as new even when \(f=0\). The right-hand side is therefore not removed; it is retained under a scaling condition ensuring that it disappears in the blow-up limit.

The parabolic counterpart, “Parabolic boundary Harnack inequalities with right-hand side” [2307.06487], extends this logic to non-divergence-form operators
\[
L u := \partial_t u - \sum_{i,j=1}^n a_{ij}(x,t)\,\partial_{ij}u
\]
in parabolic flat Lipschitz domains. The crucial scaling requirement is \(q>n+2\) for \(f\in L^q\), because under parabolic rescaling the norm of the right-hand side decays like \(r^{2-(n+2)/q}\). Under small flatness and scale-invariant smallness of the right-hand side, the quotient \(u/v\) lies in \(C^{0,\gamma}\) up to the boundary for \(\gamma<\min\{\alpha_0,1-(n+2)/q\}\). For the heat equation, the paper further obtains \(C^{1-\varepsilon}\) regularity of the quotient. The endpoint \(q=n+2\) is critical and fails in this framework. In both elliptic and parabolic settings, the governing idea is that admissible right-hand-side terms are those that become negligible after blow-up, restoring homogeneous boundary comparison at the limiting scale.

A different but related right-hand-side dependence appears in “Convexity of Second-Order Cone Program in the Right Hand Side Parameter” [2208.13941]. For the SOCP
\[
\min_x \; c^\top x \quad \text{s.t.} \quad A x=b,\; x\in \mathcal{K},
\]
the paper studies the value function
\[
v(b):=\inf\{c^\top x:Ax=b,\;x\in\mathcal{K}\}.
\]
Under strong duality, \(v(b)\) equals the support function of the dual feasible set \(\mathcal{Y}=\{y:c-A^\top y\in \mathcal{K}^*\}\):
\[
v(b)=\sup_{y\in \mathcal{Y}} b^\top y.
\]
Because a support function is convex, the optimum value is convex in the right-hand-side parameter \(b\). The paper also records the associated subgradient characterization: any dual optimal solution \(y^*\) is a subgradient of \(v\) at \(b\), and boundedness of \(\mathcal{Y}\) yields global Lipschitz continuity. Here the right-hand side is not a perturbative forcing term but the parametric variable of the optimization problem itself.

## 6. Right action in rank-metric coding

In coding theory, the phrase reappears in the title “Right-hand side decoding of Gabidulin code and applications” [2112.07322]. Gabidulin codes are rank-metric analogues of Reed–Solomon codes built from linearized \(q\)-polynomials. The decoding problem is formulated for a received word \(\mathbf{y}=\mathbf{c}+\mathbf{e}\), where \(\mathbf{c}\) is a codeword and \(\mathbf{e}\) has rank \(t\). The distinctive feature of the algorithm is that it acts on codewords on the right. After interpolating the received word by a \(q\)-polynomial \(Y=C+E\), the decoder seeks a monic right annihilator \(A\) of \(q\)-degree at most \(t\) such that
\[
E\circ A \equiv 0 \mod (X^{[m]}-X).
\]
This contrasts with Loidreau’s left-action algorithm, which searches for a left annihilator.

The resulting key equations are first nonlinear and then linearized by setting \(N=C\circ A\). For full length \(n=m\), correctness holds up to the unique decoding radius
\[
t\le \left\lfloor \frac{n-k}{2}\right\rfloor.
\]
For non-full length \(n<m\), the paper constructs a \(q\)-polynomial \(G\) with \(\mathrm{Im}(G)=\mathrm{Span}_{\mathbb{F}_q}(g_1,\dots,g_n)\), reducing the problem to a full-length instance after composition by \(G\). The same right-action viewpoint extends naturally to interleaved Gabidulin codes. In that setting, a common right annihilator \(A\) yields a joint linear system with \(un\) equations and \((u+1)t+uk+1\) unknowns, giving the familiar beyond-half-distance threshold
\[
t \le \left\lfloor \frac{un-uk-1}{u+1}\right\rfloor \approx \left\lfloor \frac{u}{u+1}(n-k)\right\rfloor
\]
when the equations are independent. The paper also uses this formulation to clarify cryptanalytic attacks and defenses, notably the effect of the \(\mathbb{F}_{q^m}\)-rank \(\vartheta\) of error rows on decoder failure. In this literature, “right-side-out” is best understood as a noncommutative algebraic orientation: composition on the right is the decisive operation.

## 7. Geometric and gravitational right-hand-side usages

In discrete geometry, “Can \(n^d + 1\) unit right \(d\)-simplices cover a right \(d\)-simplex with shortest side \(n+\epsilon\)?” [1711.08497] uses “right” in the geometric sense of right simplices. The title poses an open covering question, but the paper proves a different theorem: if \(0<\epsilon\le 1/(n+2)\), then the right \(d\)-simplex \(S_{n+\epsilon}\) can be covered by
\[
(n+1)^d + (n-1)^d - n^d
\]
unit right \(d\)-simplices. The construction splits \(S_{n+\epsilon}\) into a top and a base. The top is a translate of \(S_{n-1}\) and is triangulated by \((n-1)^d\) unit right simplices. The base is covered by \((n+1)^d-n^d\) simplices obtained from the base triangulation of \(S_{n+1}\) by a squeeze-and-shift procedure. In dimension \(d=2\), the count becomes \(n^2+2\), recovering the Conway–Soifer number after an affine equivalence to equilateral triangles. The paper does not resolve whether \(n^d+1\) pieces suffice in general.

In gravitation, “The Right Side of Einstein’s Equation” [1401.5513] takes the right-hand side literally as the source term \(T_{\mu\nu}\) in
\[
G_{\mu\nu}+\Lambda g_{\mu\nu}=8\pi G\,T_{\mu\nu}.
\]
The paper derives modified nonrelativistic and relativistic stress tensors from kinetic theory without the Chapman–Enskog truncation, retaining process-dependent terms proportional to the mean flight time \(\epsilon\). In the relativistic ultraradiative case, the first-order closure produces an effective bulk contribution \(\Pi = -(\epsilon/3)Q\), where \(Q=\dot E + (E+P)\theta\). Applied to isotropic cosmology, this yields the factorized condition
\[
Q(1-\epsilon H)=0.
\]
One branch is the standard Friedmann branch \(Q=0\); the other is a new “temporal shock” branch \(\epsilon H=1\), in which the expansion time scale is of order the mean flight time. In the isotropic setting, entropy production is then
\[
\nabla_\mu s^\mu = \epsilon H\,Q/T,
\]
which vanishes on the Friedmann branch and reduces to \(Q/T\) on the temporal-shock branch. Here the right-hand side is not merely a bookkeeping location in an equation; it is the target of a first-principles rederivation intended to alter cosmological dynamics.

Taken together, these geometric and gravitational usages show that the “right” in Right-Side-Out can refer to a geometric class, a source term, or an exterior state. The common thread is operational asymmetry: a privileged side, orientation, or right-hand-side structure is singled out and then made analytically consequential.

Source: https://www.emergentmind.com/topics/right-side-out