---
title: 'Paired Reversal: A Cross-Domain Framework'
url: https://www.emergentmind.com/topics/paired-reversal
type: topic
---

# Paired Reversal: A Cross-Domain Framework

Paired reversal is not presented in the cited literature as a single field-independent formalism. Taken together, these works suggest an umbrella notion in which an object, update rule, or physical configuration is explicitly coupled to its reverse, or a reversal applied to one component is compensated, averaged, or replayed by a partner operation. In combinatorics, this coupling is realized by comparing a permutation with its reverse under the RSK recording tableau [2108.08657]. In stochastic optimization, it is formalized as the symmetrized epoch map $\bar T_\pi=\tfrac12(T_\pi+T_{\mathrm{Rev}(\pi)})$ for without-replacement SGD [2604.00260]. In correlated lattice systems, it denotes a $\pi$-phase reversal of charge order across stripes formed by paired degrees of freedom [2402.17305]. In quantum optics, it denotes the nonlocal undoing of a partial-collapse measurement on one photon by acting on its entangled partner [1012.4597]. In stack sorting and reversible programming, it appears as a forward pass coupled to a reverse pass or inverse program [1808.04199] [1709.00828].

## 1. Cross-domain structure

The cited literature does not supply a universal definition of paired reversal. Instead, it exhibits a recurring structural pattern: a forward object or process is paired with its reversal, and the pair is used to reveal invariants, cancel order effects, recover a prior state, or complete a multi-pass computation.

| Domain | Paired objects | Role of reversal |
|---|---|---|
| RSK combinatorics | $w$ and $w^r$ | Common recording tableau |
| SGD shuffling | $\pi$ and $\mathrm{Rev}(\pi)$ | Symmetrized epoch map |
| Correlated matter | Stripe domains across a wall | $\pi$-phase charge reversal |
| Quantum optics | Two entangled photons | Nonlocal measurement reversal |
| Stack sorting | Leftovers and their reverse | Additional sorting pass |
| Program semantics | Forward and inverse programs | State restoration |

A notable distinction is that only the SGD work gives a direct formal definition of paired reversal as an operator on epoch maps [2604.00260]. In the other cases, the phrase functions as a descriptive or interpretive label for a reversal-coupled construction. This suggests that paired reversal is better treated as a structural motif than as a single canonical method.

## 2. RSK fixed points under reversal

In the RSK setting, paired reversal concerns permutations whose reverse shares the same recording tableau. For a permutation $w=w_1\cdots w_n\in\mathfrak{S}_n$, the reverse is
\[
w^r=w_n\cdots w_1.
\]
Writing $R_n:=\{\,w\in\mathfrak{S}_n:Q(w)=Q(w^r)\,\}$, the main enumerative result is
\[
|R_n|=\begin{cases}
2^{\frac{n-1}{2}}\binom{n-1}{\frac{n-1}{2}}, & n \text{ odd},\\[4pt]
0, & n \text{ even}.
\end{cases}
\]
The parity split is forced by structure: if $Q(w)=Q(w^r)$, then $Q(w)$ must have symmetric hook shape, and such a hook exists only for odd $n$ [2108.08657].

For odd $n$, the relevant shape is
\[
\lambda(n)=\left(\frac{n+1}{2},1^{\frac{n-1}{2}}\right).
\]
The structural theorem identifies the admissible recording tableaux as exactly those symmetric hook tableaux fixed by evacuation-transpose. With
\[
M_\lambda:=\{\,P\in \mathrm{SYT}(\lambda):\mathsf{E}(P)^T=P\,\},
\]
Lemma 4.6 gives the explicit criterion: for every $i\ge 2$, if $i$ lies in the first row of $P$, then $n-i+2$ lies in the first column. Equivalently, row and column entries are paired by the involution
\[
i\longleftrightarrow n-i+2.
\]

The count factorizes into a tableau count and an RSK-fiber count. First, for odd $n$,
\[
|M_\lambda|=2^{\frac{n-1}{2}},
\]
because one chooses row-or-column placement for the integers $2,3,\dots,\frac{n+1}{2}$, after which the locations of their partners are forced. Second, for each fixed $Q\in M_\lambda$, the number of permutations with $Q(w)=Q$ is $f^\lambda$, and for a hook shape $\lambda=(k,1^{n-k})$ one has
\[
f^\lambda=\binom{n-1}{k-1}.
\]
Substituting $k=\frac{n+1}{2}$ yields the closed formula above [2108.08657].

The same phenomenon can be expressed through standard RSK symmetries. Since
\[
Q(w^r)=\mathsf{E}(Q(w))^T,
\]
the condition $Q(w)=Q(w^r)$ is equivalent to
\[
Q(w)=\mathsf{E}(Q(w))^T.
\]
Paired reversal therefore picks out fixed points of evacuation followed by transpose, and the paper shows that these fixed points are rigidly confined to symmetric hooks with the row-column pairing property [2108.08657].

## 3. Symmetrized epoch maps in without-replacement SGD

In finite-sum optimization,
\[
F(w):=\frac{1}{n}\sum_{i=1}^n f_i(w),
\]
an epoch under permutation $\pi$ applies the updates
\[
w_t=w_{t-1}-\gamma\,\nabla f_{\pi(t)}(w_{t-1}),\qquad t=1,\dots,n,
\]
and defines the epoch map $T_\pi(w):=w_n$. The reversal operator is
\[
\mathrm{Rev}(\pi)=(\pi(n),\dots,\pi(1)).
\]
The paper formalizes paired reversal through the averaged epoch map
\[
\bar T_\pi(w):=\frac12\bigl(T_\pi(w)+T_{\mathrm{Rev}(\pi)}(w)\bigr),
\]
which is analyzed as a symmetrized alternative to a single ordered pass [2604.00260].

Under the assumptions that each $f_i$ is twice continuously differentiable and satisfies the bounds
\[
\|\nabla f_i(x)-\nabla f_i(y)\|\le L\|x-y\|,\qquad
\|\nabla f_i(x)\|\le G,\qquad
\|\nabla^2 f_i(x)-\nabla^2 f_i(y)\|\le \rho\|x-y\|,
\]
Lemma 3.4 gives the expansion
\[
T_\pi(w)=w-\gamma\sum_{i=1}^n g_i+\gamma^2\sum_{1\le s<t\le n}H_{\pi(t)}g_{\pi(s)}+R_\pi(w),
\]
with remainder
\[
\|R_\pi(w)\|\le C_{\mathrm{rem}}\gamma^3n^3,\qquad
C_{\mathrm{rem}}:=\frac{\rho G^2}{2}+\frac{L^2G}{2},
\]
where $g_i=\nabla f_i(w)$ and $H_i=\nabla^2 f_i(w)$. The order-sensitive term is
\[
B_\pi(w):=\sum_{1\le s<t\le n}H_{\pi(t)}g_{\pi(s)}.
\]

The paired-reversal identity
\[
B_\pi+B_{\mathrm{Rev}(\pi)}=\sum_{i\neq j}H_ig_j
\]
removes the permutation dependence at second order. Theorem 3.5 therefore yields
\[
\bar T_\pi(w)
=
w-\gamma\sum_{i=1}^n g_i
+\frac{\gamma^2}{2}\sum_{i\neq j}H_ig_j
+\bar R_\pi(w),
\]
where the second-order term is independent of $\pi$ and
\[
\|\bar T_\pi(w)-\bar T_{\pi'}(w)\|\le 2C_{\mathrm{rem}}\gamma^3n^3
\]
for any $\pi,\pi'$. By contrast, without symmetrization Theorem 3.7 gives
\[
\|T_\pi(w)-T_{\pi'}(w)\|
\le LG\,\gamma^2\,n(n-1)+2C_{\mathrm{rem}}\gamma^3n^3,
\]
and Proposition 3.8 shows that the $\Theta(\gamma^2)$ dependence is tight in general [2604.00260].

The same cancellation improves permutation variance. For standard epoch maps,
\[
\mathrm{Var}_\pi(T_\pi(w))
\le 2\gamma^4\,\mathrm{Var}_\pi(B_\pi(w))
+2C_{\mathrm{rem}}^2\gamma^6n^6,
\]
whereas
\[
\mathrm{Var}_\pi(\bar T_\pi(w))
\le C_{\mathrm{rem}}^2\gamma^6n^6.
\]
Thus paired reversal eliminates the leading permutation-random second-order contribution and reduces order sensitivity from quadratic to cubic in $\gamma$, while reducing permutation variance from a $\gamma^4$ scale to a $\gamma^6$ scale [2604.00260].

The paper analyzes this mechanism separately from block reshuffling. Block reshuffling reduces prefix-gradient variance constants through the decomposition
\[
\sigma_{\mathrm{ind}}^2(w)=\sigma_{\mathrm{within}}^2(w)+\sigma_{\mathrm{blk}}^2(w),
\]
whereas paired reversal targets order sensitivity by symmetrizing the epoch map itself. The LLM-discovered APR algorithm combines block reshuffling with periodic one-sided reversals, but the theoretical paired-reversal object remains the averaged map $\bar T_\pi$ [2604.00260].

## 4. Charge $\pi$-phase shifts in paired lattice systems

In isotropically paired lattice systems, paired reversal refers to a reversal of charge order carried by paired degrees of freedom across stripe domain walls. The paper studies two settings: a hard-core boson model representing tightly bound local Cooper pairs, and the attractive Hubbard model at finite $|U|$, where pairs remain $s$-wave but are spatially extended [2402.17305].

The hard-core boson Hamiltonian is
\[
\hat {\cal H}_{\rm hcb} =
-J\sum_{\langle i,j \rangle } (\hat b_i^\dagger \hat b_j + \hat b_j^\dagger \hat b_i)
+V \sum_{\langle i,j \rangle } \hat n_i' \hat n_j'
-\mu \sum_i \hat n_i',
\]
with hard-core constraint $\hat b_i^{\dagger 2}=\hat b_i^2=0$. The attractive Hubbard model is
\[
\hat {\cal H}
=
-t\sum_{\langle i,j\rangle,\sigma} \hat c_{i\sigma}^\dagger \hat c_{j\sigma}
+U\sum_i \hat n_{i\uparrow}\hat n_{i\downarrow}
-\mu\sum_{i\sigma}\hat n_{i\sigma},
\qquad U<0.
\]
In the strong-coupling mapping, local Cooper pairs become hard-core bosons with repulsive interactions of order $t^2/|U|$ [2402.17305].

Stripe order is diagnosed through the density structure factor
\[
S(\mathbf{k})=
\frac{1}{L_xL_y}\sum_{i,j} e^{i\mathbf{k}\cdot(\mathbf{r}_i-\mathbf{r}_j)}
\big\langle (\hat n_i-\langle \hat n_i\rangle)(\hat n_j-\langle \hat n_j\rangle)\big\rangle.
\]
A charge $\pi$-phase shift means that the checkerboard sublattice preference reverses across a stripe. In the bosonic representation this is visualized by the staggered hole density
\[
(-1)^{i_x+i_y}\Big(\tfrac12-\langle \hat n_i'\rangle\Big),
\]
which changes sign across stripe lines, and by density correlations whose checkerboard staggering reverses at stripe positions. In momentum space, the $\pi$ shift is reflected by peaks of $S(k_x,\pi)$ at incommensurate
\[
k_x=\pi(1\pm 2\delta),
\]
rather than at $(\pi,\pi)$ [2402.17305].

The principal physical conclusion is a distinction between static and fluctuating paired reversal. In the hard-core boson ladders, once stripes are formed, either via external pinning or spontaneously, a sublattice reversal of charge ordering occurs and suppresses the superfluid weight. The superfluid stiffness is
\[
\rho_s=\frac{\langle W_x^2+W_y^2\rangle}{2\beta J},
\]
and the data show that when static stripes with charge $\pi$-phase shifts become robust, $\rho_s\to 0$. By contrast, in the attractive Hubbard model with finite interactions, AFQMC shows evidence of fluctuating stripes that coexist with superfluidity, as measured by the pair structure factor
\[
P_s=\frac{1}{L_xL_y}\sum_{i,j}\langle \hat\Delta_i \hat\Delta_j^\dagger\rangle,
\qquad
\hat\Delta_i=c_{i\uparrow}c_{i\downarrow}.
\]
The paper summarizes this as the incompatibility of static stripes with pairing for local pairs, unlike the case of fluctuating ones [2402.17305].

Within this usage, paired reversal does not mean reversing a trajectory or permutation. It denotes a domain-wall-induced sign reversal in the charge order of paired objects. The paired degrees of freedom are the carriers of the order parameter whose preferred sublattice occupancy flips across the stripe.

## 5. Nonlocal reversal in entangled quantum pairs

In quantum measurement theory, paired reversal is the probabilistic undoing of a partial-collapse measurement on one member of an entangled pair by acting either on the same particle or on its partner. The experiment begins from the Bell state
\[
|\psi\rangle=\frac{1}{\sqrt{2}}\bigl(|HH\rangle+|VV\rangle\bigr)_{AB},
\]
prepared with measured concurrence $0.95\pm 0.02$ [1012.4597].

The no-click branch of the partial-collapse measurement is represented by
\[
P_M=
\begin{pmatrix}
1&0\\
0&\sqrt{1-p}
\end{pmatrix},
\]
with $p\in[0,1]$ the partial-collapse strength. For an arbitrary single-qubit state $|\phi\rangle=\alpha|H\rangle+\beta|V\rangle$, the normalized post-measurement state is
\[
|\tilde\phi'\rangle
=
\frac{\alpha |H\rangle+\beta\sqrt{1-p}\,|V\rangle}
{\sqrt{|\alpha|^2+(1-p)|\beta|^2}}.
\]
A reversal operation is constructed from
\[
P_M'=
\begin{pmatrix}
\sqrt{1-p}&0\\
0&1
\end{pmatrix},
\qquad
R_M=\frac{1}{\sqrt{1-p}}P_M',
\]
so that
\[
R_M P_M=I
\]
on the relevant conditioned subspace [1012.4597].

Applying $P_M$ on photon $A$ transforms the Bell state into
\[
|\psi_1\rangle
=
\sqrt{\frac{1}{2-p}}
\left(|HH\rangle+\sqrt{1-p}\,|VV\rangle\right),
\]
whose concurrence is
\[
C(|\psi_1\rangle)=\frac{2\sqrt{1-p}}{2-p}.
\]
The entanglement therefore decreases monotonically with $p$. The central paired-reversal result is that the original entangled state can be restored not only by a local reversal on photon $A$, but also by a nonlocal reversal on photon $B$. In the latter protocol, $(I\otimes R_M)(P_M\otimes I)$ is effectively proportional to the identity on the entangled subspace, so the partial collapse on $A$ is undone by acting only on its partner [1012.4597].

The experiment implements both operations with displaced Sagnac interferometers and characterizes them by single-qubit quantum process tomography and two-qubit state tomography. The reported reversal process fidelities are above $0.93\pm 0.01$ across the full range of $p$. After local reversal, the worst-case recovered concurrence is $0.92\pm 0.03$; after nonlocal reversal, concurrence is approximately $0.9$ for all tested $p$. For a recovered state at $p=0.5$, the measured CHSH parameter is
\[
S=2.538\pm 0.035,
\]
violating the local-realistic bound by more than $15$ standard deviations [1012.4597].

Here paired reversal is literal: a disturbance introduced on one subsystem of an entangled pair is reversed by a matched operation on the other subsystem. The pairing is supplied by entanglement, and the reversal is conditional rather than deterministic, since the protocol is built from post-selected weak-measurement branches.

## 6. Multi-pass and semantic reversibility

In stack-sorting theory, paired reversal appears as a two-pass configuration in which the leftovers from a first pass are reversed and fed through the stack again. The model reads a permutation left to right, pushes every entry onto a stack, and allows popping only when the stack top is the next required output value. After a pass, if the stack contains $s_1,\dots,s_m$ from bottom to top, the next input is
\[
s_m,s_{m-1},\dots,s_1.
\]
The rev-tier $t_{\mathrm{rev}}(\pi)$ is the number of times the stack contents must be sent back to the input with reversal before $\pi$ is fully sorted. A permutation is $(t+1)$-reverse-pass sortable if it can be sorted in at most $t+1$ passes [1808.04199].

The key structural theorem states that $t_{\mathrm{rev}}(\pi)$ equals the maximum length of an alternating sequence of separated pairs
\[
(i_1,i_1+1),\dots,(i_t,i_t+1),
\]
with $i_1<\cdots<i_t$, starting with a down separated pair and alternating in orientation. This yields an exact basis description for the two-pass case:
\[
\text{2-reverse-pass sortable permutations}=Av(2413,2431,23154).
\]
More generally, all classes of $(t+1)$-reverse-pass sortable permutations are finitely based, and the maximal rev-tier for permutations of length $n$ is $n-2$. The permutations of length $n$ with rev-tier $n-2$ form a new Entringer family, in bijection with alternating permutations of length $n-1$ [1808.04199].

In program semantics, reversal is realized by pairing a forward execution with a generated inverse program. The sequential framework transforms an original program $P$ into an augmented program $\mathrm{aug}(P)$ and an inverse program $\mathrm{inv}(P)$. Augmentation leaves the ordinary store $\sigma$ unchanged relative to the original program but records reversal information in an auxiliary store $\delta$; inversion reads $\delta$ to restore both $\sigma$ and $\delta$ to their initial states. The core correctness statements are: if
\[
(P,\sigma,\delta)\rightarrow^*(\texttt{skip},\sigma',\delta),
\]
then
\[
(\mathrm{aug}(P),\sigma,\delta)\rightarrow^*(\texttt{skip},\sigma',\delta')
\]
for some $\delta'$, and
\[
(\mathrm{inv}(P),\sigma',\delta')\rightarrow^*(\texttt{skip},\sigma,\delta)
\]
[1709.00828].

For non-communicating parallelism, the extension replaces plain augmentation by annotation with statement identifiers. Forward execution records fresh IDs via `next()`, while reverse execution uses `previous()` and the recorded ID stacks to replay the exact interleaving in reverse. This addresses the fact that naive source-order inversion is unsound under interleaving. In this setting, paired reversal is not an average or a symmetry condition; it is a forward-and-inverse program pair whose semantics guarantee exact rollback of the executed trace [1709.00828].

Taken together, these works suggest that paired reversal is a family resemblance concept rather than a single theorem schema. Its recurring content is the deliberate coupling of a process with its reverse so that symmetry becomes analyzable, second-order order effects cancel, an entangled disturbance is undone, stripe domains acquire antiphase structure, leftovers are made sortable by another pass, or an imperative execution is rendered exactly invertible.

Source: https://www.emergentmind.com/topics/paired-reversal