Elizalde constructed a bijection \phi from the cyclic permutations \pi\in S_{n+1} to the symmetric group S_n satisfying \operatorname{Des}(\pi)\cap \{1,2,\ldots,n-1\}=\operatorname{Des}(\phi(\pi)). We give a corresponding result on the signed symmetric group B_n by constructing a function \Phi from the cyclic signed permutations \pi\in B_{n+1} to B_n satisfying \operatorname{Des}(\pi)\cap \{0,1,\ldots,n-1\}=\operatorname{Des}(\Phi(\pi)). Moreover, letting D_{n+1}\subseteq B_{n+1} be the subgroup consisting of signed permutations with an even number of sign changes, we show that the restriction of \Phi to the cyclic signed permutations in D_{n+1} or its complement is a bijection. Our function \Phi reduces to Elizalde’s original bijection \phi under the natural identification of the symmetric groups as subgroups of the signed symmetric groups. One application of our results is asymptotic normality of the descent and flag major index statistics on the cyclic signed permutations in B_{n} and D_n.
[elizalde, Theorem 1] For every positive integer n, there is a bijection \phi:\operatorname{\mathfrak{C}}_{S,n+1}\to S_n satisfying
for every \pi\in \operatorname{\mathfrak{C}}_{S,n+1}.
For every positive integer n, there exists a function \Phi:\operatorname{\mathfrak{C}}_{B,n+1}\to B_n with the following properties:
for all \pi\in \operatorname{\mathfrak{C}}_{B,n+1}, we have \operatorname{Des}(\pi)\cap \{0,1,\ldots,n-1\}=\operatorname{Des}(\Phi(\pi)), and
the restriction of \Phi to \operatorname{\mathfrak{C}}_{D,n+1} or \overline{\operatorname{\mathfrak{C}}_{D,n+1}} is a bijection.
For any positive integer n and I\subseteq \{0,1,\ldots,n-1\}, the following are equal:
the number of \pi\in B_n with \operatorname{Des}(\pi)=I,
the number of \pi\in \operatorname{\mathfrak{C}}_{D,n+1} with \operatorname{Des}(\pi)\cap \{0,1,\ldots,n-1\}=I, and
the number of \pi\in \overline{\operatorname{\mathfrak{C}}_{D,n+1}} with \operatorname{Des}(\pi)\cap \{0,1,\ldots,n-1\}=I.
Let (X_n)_{n\geq 1} be the random variables corresponding to either the descent or flag major index statistic on (\operatorname{\mathfrak{C}}_{B,n})_{n\geq 1}, (\operatorname{\mathfrak{C}}_{D,n})_{n\geq 1}, or (\overline{\operatorname{\mathfrak{C}}_{D,n}})_{n\geq 1} with means (\mu_n)_{n\geq 1} and variances (\sigma_n^2)_{n\geq 1}. The standardized random variable (X_n-\mu_n)/\sigma_n converges in distribution to the standard normal distribution.
Consider the signed permutation \sigma on [\pm 6] defined by \sigma(1)=-3, \sigma(2)=1, \sigma(3)=2, \sigma(4)=-5, \sigma(5)=-4, and \sigma(6)=6. Its one-line notation is \sigma=[-3,1,2,-5,-4,6]. From the property \sigma(-i)=-\sigma(i) for i\in [n], we find that \sigma(-1)=3, \sigma(-2)=-1, \sigma(-3)=-2, \sigma(-4)=5, \sigma(-5)=4, and \sigma(-6)=-6.
Let \pi=[1,-3,-2,5,6,4]\in B_6. Letting \sigma\in B_6 be the permutation from Example 2.1, we have that \pi \sigma = [2,1,-3,-6,-5,4]. Observe that \pi\in D_6, while \sigma,\pi\sigma\notin D_6.
The permutation \sigma from Example 2.1 can be expressed as
Observe that the set of elements that appear in the cycle notation and one-line notation coincide, and we will use this fact freely throughout our work.
Consider \sigma=(-3,2,1)(-5,-4)(6) from Example 2.3. Rewriting each cycle so that the largest element is first, we obtain \sigma=(2,1,-3)(-4,-5)(6). Rearranging cycles so that first elements are in increasing order, we obtain the canonical cycle notation \sigma=(-4,-5)(2,1,-3)(6).
For the permutation \sigma=[-3,1,2,-5,-4,6] from Example 2.1, we have that \operatorname{Des}(\sigma)=\{0,3\} and \operatorname{Neg}(\sigma)=\{1,3,4\}. Using these, we calculate \operatorname{des}(\sigma)=2, \operatorname{maj}(\sigma)=3, \operatorname{neg}(\sigma)=3, and \operatorname{fmaj}(\sigma)=9.
Let (X_n)_{n\geq 1} be a sequence of real-valued random variables, and let X be a real-valued random variable. Define F_n(x)=P(X_n\leq x) and F(x)=P(X\leq x) to be their respective cumulative distribution functions.
X_n converges in distribution to X, denoted X_n\xrightarrow{d} X, if F_n(x)\to F(x) for all x that are continuity points of F.
X_n converges in probability to X, denoted X_n\xrightarrow{p} X, if for all \epsilon>0, we have that
It is well known that convergence in probability implies convergence in distribution.
Let (X_n)_{n\geq 1} and (Y_n)_{n\geq 1} be sequences of random variables such that X_n\xrightarrow{d} X and Y_n\xrightarrow{p} c, where c\in \mathbb{R} is a constant. Then X_n+Y_n\xrightarrow{d} X+c and X_nY_n\xrightarrow{d} X\cdot c.
[CM2012, Theorems 3.1 and 3.4] Let X_n be the random variable corresponding to the descent statistic on elements of B_n. Then X_n has mean \mu_n=n/2 and variance \sigma_n^2=(n+1)/12. Furthermore, as n\to\infty, the standardized random variable (X_n-\mu_n)/\sigma_n converges in distribution to \mathcal{N}(0,1).
[CM2012, Theorems 4.1 and 4.3] Let X_n be the random variable corresponding to the flag major index statistic on elements of B_n. Then X_n has mean \mu_n=n^2/4 and variance \sigma_n^2=(4n^3+6n^2-n)/36. Furthermore, as n\to\infty, the standardized random variable (X_n-\mu_n)/\sigma_n converges in distribution to \mathcal{N}(0,1).
The original statements of Theorem 2.8, Theorem 2.9 given in [CM2012] involve the more general setting of colored permutation groups. These are wreath products of the form S_{n,r}=\mathbb{Z}_r\wr S_n, where \mathbb{Z}_r is the cyclic group of order r. They contain B_n\cong S_{n,2} as a special case. See ? for further details.
While the descent statistic on S_{n,2} does not align with our descent statistic on B_n under the usual isomorphism between these groups, results of Steingrímsson [steingrim94] and Brenti [brenti94] show that the distributions of these statistics are the same on S_{n,2}\cong B_n. One can also show that the distributions of the two different flag major index statistics align, e.g., see [des_fmaj, Remark 4.8]. Hence, the current statements in Theorem 2.8, Theorem 2.9 involving these statistics on B_n is appropriate.
Let X_n be the random variable corresponding to the descent statistic on \operatorname{\mathfrak{C}}_{B,n},\operatorname{\mathfrak{C}}_{D,n}, or \overline{\operatorname{\mathfrak{C}}_{D,n}}, and let Y_n be the corresponding random variable on B_n. If n\geq 5, then the mean and variance of X_n are the same as the mean and variance of Y_n, respectively.
Let X_n be the random variable corresponding to the flag major index statistic on \operatorname{\mathfrak{C}}_{B,n},\operatorname{\mathfrak{C}}_{D,n}, or \overline{\operatorname{\mathfrak{C}}_{D,n}}, and let Y_n be the corresponding random variable on B_n. If n\geq 5, then the mean and variance of X_n are the same as the mean and variance of Y_n, respectively.
Let \pi\in \operatorname{\mathfrak{C}}_{B,n+1} and \sigma\in B_n. For brevity, we will use the notation
where \Delta denotes the symmetric difference of two sets. Using this, define for any nonnegative integers x and y the Boolean function
In a cycle notation for \pi\in \operatorname{\mathfrak{C}}_{B,n+1}, an element x\in [\pm n] is a left-to-right maximum if it is larger than every element that appears to its left.
Define \phi: \operatorname{\mathfrak{C}}_{B,n+1}^+\to B_n to be the function mapping each \pi\in \operatorname{\mathfrak{C}}_{B,n+1}^+ to its output from Algorithm ?.
Consider the signed permutation
Its descent set is \{2,4,6\}. After finding the left-to-right maxima of the cycle notation, Algorithm ? begins with
In the first iteration of the for loop in line ?, we see that P_{\pi,\sigma}(4,5) is True and P_{\pi,\sigma}(4,3) is False. Then the algorithm chooses \epsilon=1, and the first replacement in the while loop on line ? results in
Since P_{\pi,\sigma}(5,6) is False, the first iteration of the for loop ends here. In the second iteration of the for loop, we see that P_{\pi,\sigma}(1,2) is False and P_{\pi,\sigma}(1,0) is always False, so no swaps occur. In the third iteration, again no swaps occur, as P_{\pi,\sigma}(2,1) and P_{\pi,\sigma}(2,3) are both False. Consequently,
Observe that \operatorname{Des}(\phi(\pi))=\{0,2,4\}, so its descents in [5] coincide with those of \pi. However, the descent 0\in \operatorname{Des}(\phi(\pi)) is introduced by the algorithm, and the descent 6\in \operatorname{Des}(\pi) cannot occur for \phi(\pi)\in B_6.
For a larger example, consider
which has descent set \{0,6,7,9,10,12\}. The algorithm begins with
In the first iteration of the for loop, we see that P_{\pi,\sigma}(4,3) is True and P_{\pi,\sigma}(4,5) is False, so the algorithm performs swaps beginning with 4 and 3 to obtain
Since P_{\pi,\sigma}(3,2) is False, the first iteration of the for loop ends here.
In the second iteration, P_{\pi,\sigma}(6,5) is True and P_{\pi,\sigma}(6,7) is False, so the algorithm performs swaps beginning with 6 and 5 to obtain
Since P_{\pi,\sigma}(5,4) is now True, the algorithm performs further swaps beginning with 5 and 4 to obtain
As P_{\pi,\sigma}(4,3) is False, the second iteration of the for loop end here.
In the third iteration, both P_{\pi,\sigma}(10,11) and P_{\pi,\sigma}(10,9) are False. Hence, the algorithm does not perform any swaps, so
This has descent set \{0,6,7,9,10\}, and this matches \operatorname{Des}(\pi) at all elements except 12, which is not a possible descent for \phi(\pi)\in B_{12}.
Consider the application of Algorithm ? on \pi\in \operatorname{\mathfrak{C}}_{B,n+1}^+, let
be the signed permutation formed at the start of the algorithm in line ?, and define
We have that \pi_{i_1}<\pi_{i_2}<\dots < \pi_{i_m}. Furthermore, \pi_{i_j} is the largest element in \sigma_1,\sigma_2,\ldots,\sigma_j for each j\in [m].
For any x\in [n]\setminus L, we have that \pi(x)=\sigma(x), and for any x\in L, we have that \pi(x)>\sigma(x).
If an element x\in [n-1] is in \operatorname{Des}_{\Delta}(\pi,\sigma), then one of the following must be true:
x\in L, x+1\notin L, \pm (x+1) appears right of \pm x in \sigma, and \pi(x)>\pi(x+1)=\sigma(x+1)>\sigma(x), or
x\notin L, x+1\in L, \pm x appears right of \pm (x+1) in \sigma, and \pi(x+1)>\pi(x)=\sigma(x)>\sigma(x+1).
In particular, \pm x and \pm (x+1) cannot be in the same cycle.
For (a), the algorithm creates the individual cycles so that left-to-right maxima are the first elements in their respective cycles. From this, we conclude \pi_{i_1}<\pi_{i_2}<\dots < \pi_{i_m}. Additionally, \pi_{i_j} is the right-most left-to-right maximum in \sigma_1,\sigma_2,\ldots,\sigma_j, so it must be larger than all elements in those cycles.
For (b), the construction of \sigma also implies that the element following any x\in [n]\setminus L in the cycle notations of \pi and \sigma coincide, so \pi(x)=\sigma(x). For each x\in L, we find the cycle \sigma_j containing \pm x as the last element and use (a) to conclude
so (b) follows.
For (c), note that if both x,x+1\notin L, then (b) implies \pi(x)=\sigma(x) and \pi(x+1)=\sigma(x+1), so x\notin \operatorname{Des}_{\Delta}(\pi,\sigma). If both x,x+1\in L, then let \sigma_j and \sigma_k be the cycles containing \pm x and \pm (x+1) in the last entry, respectively. Using (a), the relative order of \pi(x)=\pi_{i_{j+1}} and \pi(x+1)=\pi_{i_{k+1}} matches that of \sigma(x)=\pi_{i_j} and \sigma(x+1)=\pi_{i_k}, so x\notin \operatorname{Des}_{\Delta}(\pi,\sigma). Combined, we conclude x\in \operatorname{Des}_{\Delta}(\pi,\sigma) requires that exactly one of x or x+1 is in L. The two possibilities given in (c) now follow from (a), (b), and casework on which of x or x+1 is in L.
Consider the application of Algorithm ? on \pi\in \operatorname{\mathfrak{C}}_{B,n+1}^+, and let
be the signed permutation at the start of Algorithm ?. Suppose x is the last element in some cycle of \sigma and \epsilon\in \{-1,1\}. If P_{\pi,\sigma}(|x|,|x+\epsilon|) is True and P_{\pi,\sigma}(|x|,|x-\epsilon|) is False, then \sigma(|x|)<\sigma(|x-\epsilon|)<\sigma(|x+\epsilon|) cannot occur.
Suppose that x is the last element in \sigma_j. Since P_{\pi,\sigma}(|x|,|x+\epsilon|) is True, Lemma 3.6(c) implies that
There are two cases for P_{\pi,\sigma}(|x|,|x-\epsilon|) being False. If \sigma(|x|)>\sigma(|x-\epsilon|) and \pi(|x|)>\pi(|x-\epsilon|), then combined with (Proof), we conclude
Alternatively, suppose that \sigma(|x|)<\sigma(|x-\epsilon|) and \pi(|x|)<\pi(|x-\epsilon|). If \pm (x-\epsilon) is not the last element of a cycle, then Lemma 3.6(b) implies \sigma(|x-\epsilon|)=\pi(|x-\epsilon|), so combined with (Proof), we conclude
If \pm (x-\epsilon) is the last element of a cycle, then Lemma 3.6(a) with \pi(|x|)<\pi(|x-\epsilon|) implies \pm (x-\epsilon) appears to the right of \pm x. Combined with (Proof), we find
In all cases, \sigma(|x|)<\sigma(|x-\epsilon|)<\sigma(|x+\epsilon|) is not possible.
The order properties of {\sigma} are the following.
The relative order of elements within each cycle \sigma_1,\ldots,\sigma_m matches the relative order within the respective cycle of \pi_1,\ldots,\pi_m. In particular, the largest element in each cycle is the first one.
The first elements of the cycles satisfy \sigma_{1,1}<\sigma_{2,1}<\dots <\sigma_{m,1}. Additionally, for every i\in [m], the first element \sigma_{i,1} is the largest element that appears in \sigma_1,\sigma_2,\ldots,\sigma_i.
Suppose x\in [\pm n] appears in the cycle notation of \sigma, and let k\geq j+1. Then \pi(|x|)>\pi_{k,1} if and only if \sigma(|x|)\geq \sigma_{k,1}, where equality holds when x\in \sigma_k is the last element. Furthermore, any x satisfying these equivalent properties has not been involved in any swaps.
If d\in [n-1] is in symmetric difference \operatorname{Des}_{\Delta}(\pi,\sigma), then exactly one of \pm d and \pm (d+1) is the last element in some {\sigma}_{j},\ldots,{\sigma}_{m} and the remaining element appears in the non-last position of some cycle to its right. Furthermore, letting x be the element that appears in a last position and \pm (x+\epsilon) with \epsilon\in \{-1,1\} be the other element, it must be that \pi(|x|)>\pi(|x+\epsilon|) and {\sigma}(|x|)<{\sigma}(|x+\epsilon|).
The order properties hold when \sigma is the initial permutation in line ? of Algorithm ?.
The algorithm starts with \sigma=\sigma_1\sigma_2\dots \sigma_m=\pi_1\pi_2\dots \pi_m. Each of the claims in the order properties then follow immediately from this or from Lemma 3.6.
Suppose the order properties hold for \sigma and no swaps are performed in this iteration of the for loop. Then the order properties hold for \sigma considered as the signed permutation at the start of iteration j+1 of the for loop.
Order properties ? and ? are independent of the for loop iteration, so these are immediate. By starting with order property ? for \sigma as the signed permutation in iteration j and specializing to k\geq j+2, we conclude order property ? for \sigma as the signed permutation in iteration j+1. For order property ?, let z\in \sigma_j be the last element. As the for loop does not perform any swaps, it must be that both P_{\pi,\sigma}(|z|,|z-1|) and P_{\pi,\sigma}(|z|,|z+1|) are False. Hence, z\in \sigma_j is not involved in any descent in \operatorname{Des}_{\Delta}(\pi,\sigma)\cap [n-1], so \sigma_j can be omitted from order property ? for \sigma as the signed permutation in iteration j. This results in order property ? when \sigma is considered as the signed permutation in iteration j+1.
The swap properties of {\sigma} are the following statements about the swaps performed on \sigma by the while loop in line ?.
All swaps occur between an element in the j-th cycle and an element in a cycle to its right.
The last elements in cycles \sigma_{j+1},\ldots,\sigma_m are not affected by swaps. Furthermore, if the first swap is between z\in \sigma_j and \pm (z+\epsilon) where \epsilon\in \{-1,1\}, then no element right of \pm (z+\epsilon) in the cycle notation of \sigma is affected by a swap.
If an element x in the cycle notation of \sigma satisfies \sigma(|x|)\geq \sigma_{j+1,1}, then x is not affected by swaps.
Suppose the last element z\in \sigma_j is swapped with \pm (z+\epsilon) where \epsilon\in \{-1,1\}. Then the following statements hold for the cumulative effects of the swaps on \operatorname{Des}_{\Delta}(\pi,\sigma)\cap [n-1].
The descent \min\{|z|,|z+\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi,\sigma) is removed from the symmetric difference.
Suppose \min\{|z|,|z-\epsilon|\}\in [n-1]. If \min\{|z|,|z-\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi,\sigma), then this is removed from the symmetric difference, and if \min\{|z|,|z-\epsilon|\}\notin \operatorname{Des}_{\Delta}(\pi,\sigma), then this is not introduced.
Suppose \min\{|z+\epsilon|,|z+2\epsilon|\}\in [n-1]. If \min\{|z+\epsilon|,|z+2\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi,\sigma), then this descent is removed. If \min\{|z+\epsilon|,|z+2\epsilon|\}\notin \operatorname{Des}_{\Delta}(\pi,\sigma) and \sigma(|z+2\epsilon|)>\sigma(|z+\epsilon|), then this descent is not introduced to the symmetric difference.
Any value in [n-1] that is not \min\{|z+\epsilon|,|z+2\epsilon|\} cannot be introduced into the symmetric difference.
Suppose the order properties hold for \sigma. Then the while loop in line ? continues until one of the following occurs:
the first element of \sigma_j is involved in a swap,
the swapped elements x and y do not have the same sign, or
the absolute values of the elements preceding x and y differ by more than 1.
Let z be the last element of \sigma_j, and let \pm(z+\epsilon) with \epsilon\in \{-1,1\} be the element that it is swapped with. A swap occurs due to P_{\pi,\sigma}(|x|,|y|) being True, where x=z and y=\pm (z+\epsilon). If x is the first element of \sigma_j, then by the construction of the while loop, it will terminate after this swap.
Otherwise, let \tilde{\sigma} be the permutation after x and y are swapped, and consider the elements x' and y' that precede the ones swapped. Order property ? for \sigma implies that P_{\pi,\sigma}(|x'|,|y'|) is False. If x and y did not have the same sign, then the elements that follow x' and y' in the cycle notations of \sigma and \tilde{\sigma} have the same relative order, so P_{\pi,\tilde{\sigma}}(|x'|,|y'|) is also False. If |x'| and |y'| differ by more than 1, then P_{\pi,\tilde{\sigma}}(|x'|,|y'|) is False from its definition. Hence, the while loop will terminate in both of these cases at \tilde{\sigma}.
In the remaining cases, x and y must be consecutive elements in [\pm n], and |x'| and |y'| must be consecutive elements in [n]. Swapping x and y in \sigma causes P_{\pi,\tilde{\sigma}}(|x'|,|y'|) to be True, and the while loop will continue. Iterating this argument, we see that the while loop will continue until one of the three stated situations occurs.
Suppose the order properties hold for \sigma. Then swap property ? holds for \sigma, and order property ? holds for \sigma'.
Order property ? for \sigma implies that the first swap in the while loop is between z, the last element of \sigma_j, and \pm(z+\epsilon) with \epsilon\in \{-1,1\}, which must be a non-last element in some \sigma_k with k>j. Consequently, swaps throughout the while loop only occur between \sigma_j and \sigma_{k}, so swap property ? holds for \sigma.
Order property ? for \sigma implies that the relative order of elements in each cycle of \sigma_1,\sigma_2,\ldots,\sigma_m matches the corresponding ones in \pi_1,\pi_2,\ldots,\pi_m. Swap property ? for \sigma implies that swaps only occur between elements in different cycles. Each of these swaps increases or decreases a single entry in each cycle by 1, so the relative order of elements within each cycle cannot change. We conclude order property ? holds for \sigma'.
Suppose the order properties hold for \sigma. If \sigma_{j+1,1}=\sigma_{j,1}+1, then swaps do not occur between \sigma_j and \sigma_{j+1}.
Assume by contradiction that \sigma_{j+1,1}=\sigma_{j,1}+1 and a swap occurs between \sigma_j and \sigma_{j+1}. The first swap is between the last element z\in \sigma_j and \pm (z+\epsilon)\in \sigma_{j+1} due to \min\{|z|,|z+\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi,\sigma). Order property ? for \sigma implies \pm(z+\epsilon) is not the last element of \sigma_{j+1} and \pi(|z|)>\pi(|z+\epsilon|). Since \pm(z+\epsilon) is not the last element of \sigma_{j+1}, order property ? for \sigma implies that \sigma(|z+\epsilon|)<\sigma_{j+1,1}=\sigma_{j,1}+1. Since \pm(z+\epsilon)\notin \sigma_j, we have that \sigma(|z+\epsilon|)\neq \sigma_{j,1}, so combined with \sigma(|z+\epsilon|)<\sigma_{j+1,1}=\sigma_{j,1}+1, we conclude \sigma(|z+\epsilon|)<\sigma_{j,1} = \sigma(|z|). This aligns with the relative order of \pi(|z+\epsilon|) and \pi(|z|), which contradicts \min\{|z|,|z+\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi,\sigma).
Suppose the order properties hold for \sigma and the while loop performs swaps between elements in \sigma_j and \sigma_{k} with k>j. If a swap involving \sigma_{k,1} occurs, then \sigma_{k,1}>0, and this swap must be with -\sigma_{k,1}+ \epsilon\in \sigma_j for some \epsilon \in \{-1,1\}.
If k>j+1, then order property ? for \sigma implies \sigma_{k,1}>\sigma_{j+1,1}\geq \sigma_{j,1}+1. If k=j+1, then Lemma 3.14 implies \sigma_{k,1}\neq \sigma_{j,1}+1, so combined with order property ? for \sigma, we similarly conclude that \sigma_{k,1}>\sigma_{j,1}+1. Order property ? for \sigma implies \sigma_{j,1} is the largest element in \sigma_j, so in both cases, we see that \sigma_{k,1}\pm 1\notin \sigma_j. As swaps can only involve pairs of elements whose absolute values differ by 1, we conclude that a swap involving \sigma_{k,1}\in \sigma_k can only occur if this is swapped with -\sigma_{k,1}+\epsilon\in \sigma_j for some \epsilon\in \{-1,1\}. Order property ? for \sigma implies that \sigma_{k,1} is larger than all elements in \sigma_j, so it must also be that \sigma_{k,1}>0 and -\sigma_{k,1}+\epsilon<0.
Suppose the order properties hold for \sigma. Then swap property ? holds for \sigma, and order property ? holds for \sigma'.
By order property ? for \sigma, the swaps start at the last element z\in \sigma_j and some non-last element \pm (z+\epsilon)\in \sigma_{k} with k>j. If the swaps terminate without a swap involving \sigma_{k,1}, then the last element in \sigma_{k} cannot be involved in a swap. Hence, we must analyze when a swap involving \sigma_{k,1} occurs. In this case, Corollary 3.15 implies that \sigma_{k,1} must swap with an element of the opposite sign, so Lemma 3.12 implies that the swaps will terminate at this step. Hence, the algorithm will not proceed into swapping the last element of \sigma_{k}. As the last element of \sigma_k is not involved in a swap, neither can any element right of \pm (z+\epsilon) in the cycle notation of \sigma, so swap property ? holds for \sigma.
We now consider order property ? for \sigma'. Order property ? holds for \sigma, so the first elements in \sigma_1,\sigma_2,\ldots,\sigma_m are in increasing order. By Lemma 3.13, swaps only occur between \sigma_j and some \sigma_{k} with k>j, so consider any swaps that affect the first elements in these cycles, which are \sigma_{j,1} and \sigma_{k,1}. If no swaps occur involving these elements, then it follows that the first elements in \sigma_1',\sigma_2',\ldots,\sigma_m' are also in increasing order. If a swap involves only \sigma_{j,1} (resp. \sigma_{k,1}), then the first element in cycle j (resp. k) changes by 1, while all other remaining first elements are unchanged. This also cannot affect the relative order between any of the first elements in the cycles. If a swap involves both \sigma_{j,1} and \sigma_{k,1}, then Corollary 3.15 implies that \sigma_{j,1} and \sigma_{k,1} have opposite signs. A swap changes these elements of opposite signs by 1 each, which again cannot affect the relative order between any of the first elements. In all cases, the first elements in \sigma_1',\sigma_2',\ldots,\sigma_m' are in increasing order, which is the first claim in order property ?. Lemma 3.13 implies that the largest element in each cycle of \sigma' is the first one. Combined, we see that for each i\in [m], \sigma_{i,1}' is the larger than any element in \sigma_1',\ldots,\sigma_i', so order property ? holds for \sigma'.
Suppose the order properties hold for \sigma, and the order and swap properties hold for all signed permutations prior to \sigma in the algorithm. Then swap property ? holds for \sigma.
Let x be an element appearing in the cycle notation of \sigma that satisfies \sigma(|x|)\geq \sigma_{j+1,1}. Order property ? for \sigma implies that x must be in one of the cycles \sigma_{j+1},\ldots,\sigma_k. Furthermore, if x is the last element in the cycle that contains it, then Lemma 3.16 implies that x cannot be involved in any swaps. Hence, it suffices to consider when x appears as a non-last element in some cycle \sigma_k, where k\geq j+1. As \sigma(|x|)=\sigma_{j+1,1} can only occur if x is the last element of \sigma_{j+1}, we can assume k>j+1, so \sigma(|x|)>\sigma_{j+1,1}. We will consider cases based on iteration of the while loop in line ?.
We first show by contradiction that x cannot be involved in the first swap of the while loop. For this to occur, order property ? for \sigma implies that x=\pm (z+\epsilon), where z is the last element of \sigma_j, \epsilon\in \{-1,1\}, and \min\{|x|,|z|\}\in \operatorname{Des}_{\Delta}(\pi,\sigma). By order property ? for \sigma, we have that
and \pi(|x|)<\pi(|z|). As \sigma(|x|)> \sigma_{j+1,1}, order property ? for \sigma implies that \pi(|x|)>\pi_{j+1,1}, so combined, we have \pi(|z|)>\pi(|x|)>\pi_{j+1,1}. However, order property ? for \sigma then implies that z cannot have been involved in swaps. As z is the last element in \sigma_j, it must also be the last element in \pi_j. Then \pi(|z|)=\pi_{j+1,1}, which contradicts \pi(|z|)>\pi(|x|)>\pi_{j+1,1}.
We next consider the case where x is not the first swap in this iteration of the while loop. For x to be involved in a swap, the element following x, which is \sigma(|x|), must be swapped first. As \sigma(|x|)>\sigma_{j+1,1}>\sigma_{j,1}, order property ? for \sigma implies that \sigma(|x|)\pm 1\notin \sigma_j. Then \sigma(|x|) can only being swapped with -\sigma(|x|)\pm 1. By Lemma 3.12, the while loop will terminate after swapping these elements of opposite signs, so x itself will not be involved in a swap. In all cases, x cannot be involved in a swap, so swap property ? holds for \sigma.
Suppose the order properties hold for \sigma, and the order and swap properties hold for all signed permutations prior to \sigma in the algorithm. Then order property ? holds for \sigma'.
We start with order property ? for \sigma, which states that for k\geq j+1, we have \pi(|x|)> \pi_{k,1} if and only if \sigma(x)\geq \sigma_{k,1}, where equality holds only when x is the last element in \sigma_k. Swap property ? for \sigma holds by Lemma 3.17, so this implies that any such x will not be involved in any swaps. If x is the last element of \sigma_k, then it is clear that \sigma'(|x|)=\sigma_{k,1}'. Otherwise, \sigma(|x|) and \sigma_{k,1} are distinct elements. A swap cannot simultaneously involve both \sigma_{k,1} and \sigma(|x|) by Lemma 3.13, as neither element is in \sigma_j. Additionally, a swap that changes only one of \sigma_{k,1} or the element following x does not affect their relative order. Consequently, we see that \sigma'(|x|)\geq \sigma'_{k,1} holds. Similar reasoning shows that swaps cannot introduce a new element x' satisfying \sigma'(|x'|)\geq\sigma'_{k,1} without swapping some x satisfying \sigma(|x|)\geq \sigma_{k,1}, so \sigma(|x|)\geq \sigma_{k,1} if and only if \sigma'(|x|)\geq \sigma'_{k,1}, where it is clear that equality can only hold when x is the last element of \sigma_k and \sigma_k'. In the case where the for loop remains in iteration j after swaps are performed, we conclude order property ? for \sigma'. In the case where the for loop proceeds to iteration j+1, then order property ? follows by specializing to k\geq j+2.
Suppose the order properties hold for \sigma, and the swap properties hold for all signed permutations prior to \sigma in the algorithm. Let z be the last element in \sigma_j, and let \pm (z+\epsilon) with \epsilon\in \{-1,1\} be the element that it is swapped with. The swaps in the algorithm remove the descent \min\{|z|,|z+\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi,\sigma).
From order property ? for \sigma, swaps begin due to \min\{|z|,|z+\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi,\sigma), where z\in \sigma_j is the last element and \pm (z+\epsilon) is a non-last element in some \sigma_{k} with k>j. The first swap interchanging z with \pm(z+\epsilon) changes the relative order of the elements following them in the cycle notation, which removes the descent \min\{|z|,|z+\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi,\sigma). Lemma 3.16 implies that the element now following \pm z in the cycle notation does not change from the later swaps. While these later swaps can involve the element following \pm (z+\epsilon), this element changes by at most 1, so the relative order of what follows z and \pm(z+\epsilon) does not change after the first swap. We conclude the descent \min\{|z|,|z+\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi,\sigma) is removed.
Suppose the order properties hold for \sigma, and the order and swap properties hold for all signed permutations prior to \sigma in the algorithm. Let z be the last element in \sigma_j, let \pm (z+\epsilon) with \epsilon\in \{-1,1\} be the element that it is swapped with, and suppose \min\{|z|,|z-\epsilon|\}\in [n-1]. If \min\{|z|,|z+\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi,\sigma) and \min\{|z|,|z-\epsilon|\}\notin \operatorname{Des}_{\Delta}(\pi,\sigma), then \sigma(|z|)<\sigma(|z-\epsilon|)<\sigma(|z+\epsilon|) cannot occur.
If \sigma is the signed permutation at the start of Algorithm ?, then this follows from Lemma 3.7. Otherwise, some swaps have occurred, and we will prove the result in this case by contradiction. Assume \sigma(|z|)<\sigma(|z-\epsilon|)<\sigma(|z+\epsilon|) holds. Since \min\{|z|,|z+\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi,\sigma) and \min\{|z|,|z-\epsilon|\}\notin \operatorname{Des}_{\Delta}(\pi,\sigma), it must be that
Now let \tilde{\sigma}=\tilde{\sigma}_1\tilde{\sigma}_2\dots \tilde{\sigma}_m be the signed permutation prior to the last sequence of swaps, so in a previous iteration of the while loop, swaps on \tilde{\sigma} resulted in \sigma. We consider two cases based on the iteration of the for loop.
First, suppose that the for loop was not in iteration j when swaps are performed on \tilde{\sigma}. Then swap property ? for permutations prior to \sigma implies that the last element of cycle j has not been involved in any swaps, and since this is z in \sigma, it must have been z throughout the algorithm. In particular, (Proof) becomes
From order property ? for \sigma, we conclude \sigma(|z-\epsilon|)>\sigma_{j+1,1}. However, combined with \sigma(|z+\epsilon|)>\sigma(|z-\epsilon|), this implies \sigma(|z+\epsilon|)>\sigma_{j+1,1}. Comparing with (?), we see that \pm (z+\epsilon) is an element contradicting order property ? for \sigma.
Alternatively, suppose the for loop was in iteration j when swaps are performed on \tilde{\sigma}. We begin by considering the positions of z, \pm (z-\epsilon), and \pm (z+\epsilon) in the cycles of \tilde{\sigma} containing them. As z is the last element of \sigma_j, this element must have been swapped into this position during the swaps on \tilde{\sigma}. By construction of the algorithm, \pm z must have been swapped with \pm (z-\epsilon) due to \min\{|z|,|z-\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi,\tilde{\sigma}). In particular, order property ? for \tilde{\sigma} implies that \pm (z-\epsilon) is the last element of \tilde{\sigma}_j and \pm z is a non-last of a cycle to the right of \tilde{\sigma}_j. Order property ? for \sigma with \min\{|z|,|z+\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi,\sigma) implies that \pm (z+\epsilon) is a non-last element in \sigma_{j+1},\ldots,\sigma_m. Then this must have been a non-last element in \tilde{\sigma}_j,\tilde{\sigma}_{j+1},\ldots,\tilde{\sigma}_{j+1} by swap properties ? and ? for \tilde{\sigma}.
Combined, we see that \pm (z-\epsilon) is the last element in \tilde{\sigma}_j, and both \pm z and \pm (z+\epsilon) are non-last elements in the cycles of \tilde{\sigma} containing them. In particular, order property ? for \tilde{\sigma} implies that \min\{|z|,|z+\epsilon|\}\notin \operatorname{Des}_{\Delta}(\pi,\tilde{\sigma}). The swaps on \tilde{\sigma} must introduce the descent \min\{|z|,|z+\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi,\sigma), so swap property ? for \tilde{\sigma} implies that \tilde{\sigma}(|z+\epsilon|)<\tilde{\sigma}(|z|). As order property ? for \tilde{\sigma} implies \tilde{\sigma}(|z-\epsilon|)<\tilde{\sigma}(|z|), we see that swapping \pm (z-\epsilon) and \pm z results in the element following \pm (z-\epsilon) being larger than the one following \pm (z+\epsilon). Lemma 3.16 implies that swaps do not affect the element following \pm (z-\epsilon) in the cycle notation, while they can only change what follows \pm (z+\epsilon) by 1. Hence, we conclude \sigma(|z+\epsilon|)<\sigma(|z-\epsilon|), and this contradicts \sigma(|z|)<\sigma(|z-\epsilon|)<\sigma(|z+\epsilon|).
Suppose the order properties hold for \sigma, and the swap properties hold for all signed permutations prior to \sigma in the algorithm. Let z be the last element in \sigma_j, let \pm (z+\epsilon) with \epsilon\in \{-1,1\} be the element that it is swapped with, and suppose \min\{|z|,|z-\epsilon|\}\in [n-1].
If \min\{|z|,|z-\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi,\sigma), then swaps remove this descent from the symmetric difference.
If \min\{|z|,|z-\epsilon|\}\notin \operatorname{Des}_{\Delta}(\pi,\sigma), then swaps do not introduce this to the symmetric difference.
We first consider (a). In this case, we have that \min\{|z|,|z+\epsilon|\} and \min\{|z|,|z-\epsilon|\} are both elements of \operatorname{Des}_{\Delta}(\pi,\sigma). Order property ? for \sigma implies that \pi(|z|)>\pi(|z+\epsilon|), \sigma(|z|)<\sigma(|z+\epsilon|), \pi(|z|)>\pi(|z-\epsilon|), and \sigma(|z|)<\sigma(|z-\epsilon|). Algorithm ? chooses to swap z with \pm (z+\epsilon) instead of \pm (z-\epsilon), and this can only occur if \sigma(|z+\epsilon|)>\sigma(|z-\epsilon|)>\sigma(|z|). Consequently, swapping z with \pm (z+\epsilon) also removes the descent \min\{|z|,|z-\epsilon|\}. Using a similar argument as in Lemma 3.19 and Lemma 3.20, the status of this descent is not affected by the remaining swaps in this iteration of the while loop in line ?.
For (b), we have two cases for \min\{|z|,|z-\epsilon|\}\notin \operatorname{Des}_{\Delta}(\pi,\sigma). The first scenario is \pi(|z|)>\pi(|z-\epsilon|) and \sigma(|z|)>\sigma(|z-\epsilon|), and in this case, order property ? implies \sigma(|z+\epsilon|)>\sigma(|z|)>\sigma(|z-\epsilon|). Then swapping z with \pm(z+\epsilon) does not introduce the descent \min\{|z|,|z-\epsilon|\} into the symmetric difference. As in earlier arguments, subsequent swaps also do not. The second scenario is \pi(|z|)<\pi(|z-\epsilon|) and \sigma(|z|)<\sigma(|z-\epsilon|). Order property ? for \sigma implies that \pi(|z|)>\pi(|z+\epsilon|) and \sigma(|z|)<\sigma(|z+\epsilon|), and combined with Lemma 3.20, we conclude that \sigma(|z|)<\sigma(|z+\epsilon|)<\sigma(|z-\epsilon|). A similar argument shows that swaps again do not introduce \min\{|z|,|z-\epsilon|\} into the symmetric difference.
Suppose the order properties hold for \sigma, and the swap properties hold for all signed permutations prior to \sigma in the algorithm. Let z be the last element in \sigma_j, let \pm (z+\epsilon) with \epsilon\in \{-1,1\} be the element that it is swapped with, and suppose \min\{|z+\epsilon|,|z+2\epsilon|\}\in [n-1].
If \min\{|z+\epsilon|,|z+2\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi,\sigma), then this descent is removed.
If \min\{|z+\epsilon|,|z+2\epsilon|\}\notin \operatorname{Des}_{\Delta}(\pi,\sigma) and \sigma(|z+2\epsilon|)>\sigma(|z+\epsilon|), then this descent is not introduced to the symmetric difference
For (a), both \min\{|z|,|z+\epsilon|\} and \min\{|z+\epsilon|,|z+2\epsilon|\} are elements of \operatorname{Des}_{\Delta}(\pi,\sigma). By order property ? for \sigma, this can only occur if \pm (z+\epsilon) is a non-last element in some cycle \sigma_k with k>j, and combined with order property ?, we also conclude that \pm (z+2\epsilon) must be the last element in some cycle in \sigma_{j+1},\ldots,\sigma_{k-1}. Additionally, the combination of these properties also implies \sigma(|z|)<\sigma(|z+2\epsilon|)<\sigma(|z+\epsilon|). The first swap between z and \pm (z+\epsilon) will remove \min\{|z+\epsilon|,|z+2\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi,\sigma). Later swaps in the while loop can only change the element following \pm (z+\epsilon) by at most 1 and do not affect the cycle containing \pm (z+2\epsilon). Consequently, \min\{|z+\epsilon|,|z+2\epsilon|\} cannot be re-introduced into the symmetric difference, so (a) follows.
For (b), order property ? for \sigma and the assumptions imply \sigma(|z|)<\sigma(|z+\epsilon|)<\sigma(|z+2\epsilon|), where \pm (z+\epsilon) appears as a non-last element in \sigma_{j+1},\ldots,\sigma_{m}. We let \tilde{\sigma} be the signed permutation obtained by swapping z and \pm (z+\epsilon), which satisfies \sigma(|z+\epsilon|)=\tilde{\sigma}(|z|). The relative orders of this element and what follows z and \pm (z+2\epsilon) in the cycle notation of \sigma is preserved by this first swap, so \tilde{\sigma}(|z+\epsilon|)<\tilde{\sigma}(|z|)<\tilde{\sigma}(|z+2\epsilon|). As \tilde{\sigma}(|z+\epsilon|)\neq \tilde{\sigma}(|z+2\epsilon|)-1, later swaps cannot simultaneously involve the elements following z+\epsilon and \pm (z+2\epsilon) in the cycle notation of \tilde{\sigma}. Consequently, the relative order of these elements is preserved, and \min\{|z+\epsilon|,|z+2\epsilon|\} is not introduced into the symmetric difference. We conclude that (b) holds.
Suppose the order properties hold for \sigma, and the swap properties hold for all signed permutations prior to \sigma in the algorithm. Then swap property ? holds for \sigma.
The first three claims of swap property ? for \sigma involve \min\{|z|,|z+\epsilon|\}, \min\{|z|,|z-\epsilon|\}, and \min\{|z+\epsilon|,|z+2\epsilon|\}, and the results for these are given in Lemma 3.19, Lemma 3.21, Lemma 3.22. For all other descents, consider the effect on the one-line notation when z and \pm (z+\epsilon) are swapped. In addition to affecting the images of |z| and |z+\epsilon|, this changes the images of |\sigma^{-1}(z)| and |\sigma^{-1}(z+\epsilon)| by 1. From the one-line notation of \sigma, this cannot introduce other descents into the symmetric difference unless |\sigma^{-1}(z)| and |\sigma^{-1}(z+\epsilon)| are consecutive elements of [n]. In this case, the algorithm then swaps the elements proceeding z and \pm(z+\epsilon) to remove this descent. Iterating this argument, swaps continue until one of the three cases in Lemma 3.12 occurs. If swaps end due to the first element of \sigma_j being swapped, then this can only affect descents that involve \pm (z+\epsilon), and these are accounted for in Lemma 3.19, Lemma 3.22. In the remaining two cases, the final swap does not introduce any descent, so we conclude swap property ? holds for \sigma.
Suppose the order properties hold for \sigma, and the swap properties hold for all signed permutations prior to \sigma in the algorithm. Then order property ? holds for \sigma'.
We begin with order property ? for \sigma, and apply the swap properties for \sigma, which hold by Lemma 3.13, Lemma 3.16, Lemma 3.17, Lemma 3.23. Swap property ? implies that \min\{|z|,|z-\epsilon|\} and \min\{|z|,|z+\epsilon|\} cannot be elements of \operatorname{Des}_{\Delta}(\pi,\sigma'), and descents other than \min\{|z+\epsilon|,|z+2\epsilon|\} cannot be introduced. We will begin by considering descents different from these three aforementioned ones and then consider possibilities for \min\{|z+\epsilon|,|z+2\epsilon|\}.
Let d\in \operatorname{Des}_{\Delta}(\pi,\sigma) be a descent that is not \min\{|z|,|z+\epsilon|\}, \min\{|z|,|z-\epsilon|\}, or \min\{|z+\epsilon|,|z+2\epsilon|\}. Using order property ? for \sigma, we let x\in \{\pm d,\pm (d+1)\} be whichever element appears as the last element of a cycle in \sigma_{j+1},\ldots,\sigma_m and y\in \{\pm d,\pm (d+1)\} be the element that appears in the non-last position of a cycle to its right. Combined with order property ?, we also have that \sigma_{j+1,1}\leq \sigma(|x|)<\sigma(|y|) while \pi(|x|)>\pi(|y|). By swap property ? for \sigma, the cycles containing x and y cannot both be affected by swaps, and by swap property ?, x and y themselves are not affected by swaps. Hence, swaps preserve the positions of x and y in the cycle notation and change the element following at most one of them by 1, resulting in \sigma'(|x|)<\sigma'(|y|). We conclude that the claims in order property ? involving d\in \operatorname{Des}_{\Delta}(\pi,\sigma') hold.
Finally, we consider \min\{|z+\epsilon|,|z+2\epsilon|\}. If \min\{|z+\epsilon|,|z+2\epsilon|\}\notin \operatorname{Des}_{\Delta}(\pi,\sigma'), then the for loop terminates iteration j, and the last element of \sigma_{j}', which is \pm (z+\epsilon), is not involved in any descents in \operatorname{Des}_{\Delta}(\pi,\sigma'). In this case, we conclude order property ? holds for \sigma'. It now suffices to consider when swaps result in \min\{|z+\epsilon|,|z+2\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi,\sigma'). By swap property ? for \sigma, this can only occur if \min\{|z+\epsilon|,|z+2\epsilon|\}\notin \operatorname{Des}_{\Delta}(\pi,\sigma) and \sigma(|z+2\epsilon|)<\sigma(|z+\epsilon|), which combined also imply \pi(|z+2\epsilon|)<\pi(|z+\epsilon|). We now rule out cases for where \pm (z+2\epsilon) can appear in \sigma'.
As \pm (z+\epsilon) is involved in the swaps on \sigma, swap property ? for \sigma shows that \sigma(|z+2\epsilon|)<\sigma(|z+\epsilon|)<\sigma_{j+1,1}. Then order property ? for \sigma implies that \pm (z+2\epsilon) cannot be the last element in \sigma_{j+1},\ldots,\sigma_m, and swap property ? for \sigma implies that swaps cannot result in \pm (z+2\epsilon) becoming the last element in \sigma_{j+1}',\ldots,\sigma_m'.
If \pm (z+2\epsilon) is in \sigma_1',\sigma_2',\ldots,\sigma_j', then Lemma 3.16 implies that \sigma'(|z+2\epsilon|)<\sigma_{j,1}'=\sigma'(|z+\epsilon|). Since this order aligns with \pi(|z+2\epsilon|)<\pi(|z+\epsilon|), this would imply \min\{|z+\epsilon|,|z+2\epsilon|\}\notin \operatorname{Des}_{\Delta}(\pi,\sigma).
Combined, we see that \min\{|z+\epsilon|,|z+2\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi,\sigma') can only occur if \pm(z+2\epsilon) is a non-last element in \sigma_{j+1}',\ldots,\sigma_m'. Furthermore, it must be that \pi(|z+2\epsilon|)<\pi(|z+\epsilon|) and \sigma'(|z+2\epsilon|)>\sigma'(|z+\epsilon|), so order property ? holds for \sigma'.
Let \sigma be the signed permutation at the start of an iteration of the for loop or the while loop in line ? of Algorithm ?. Then the order properties and swap properties hold for \sigma.
We prove this using strong induction. Lemma 3.6 implies that the order properties hold for the permutation \sigma at the start of the algorithm. Additionally, the swap properties vacuously hold for all permutations prior to \sigma in the algorithm. Assuming the order properties hold for \sigma, and the order and swap properties held prior to \sigma, Lemma 3.10, Lemma 3.13, Lemma 3.16, Lemma 3.17, Lemma 3.18, Lemma 3.23, Lemma 3.24 imply that the swap properties hold for \sigma and the order properties will hold for the signed permutation at the next iteration of the for or while loop. Hence, by strong induction, these properties hold throughout the algorithm.
Let \pi\in \operatorname{\mathfrak{C}}_{B,n+1}^+. Then \operatorname{Des}(\pi)\cap [n-1]=\operatorname{Des}(\phi(\pi)).
Let \sigma be the signed permutation in iteration m of the for loop. By Theorem 3.25, order property ? holds for \sigma. Then d\in \operatorname{Des}_{\Delta}(\pi,\sigma)\cap [n-1] can only exist if \pm d is the last element of \sigma_m and \pm (d+1) is in a cycle to its right, but there are no such cycles. We conclude \operatorname{Des}_{\Delta}(\pi,\sigma)\cap [n-1]=\emptyset, and this last iteration of the for loop performs no swaps. From this, we see that \phi(\pi)=\sigma and \operatorname{Des}(\pi)\cap [n-1]=\operatorname{Des}(\phi(\pi)).
Let \sigma be the signed permutation at the start of an iteration of the for loop or the while loop in line ? of Algorithm ?. Then \sigma is in canonical cycle notation.
Observe that the canonical cycle notation properties are given by order property ?. Hence, this follows immediately from Theorem 3.25.
If 0\in \operatorname{Des}(\pi), then 0\in \operatorname{Des}(\sigma). In this case, we have that \pi(1)<-1 and \sigma(1)<-1.
We will use a proof by induction, beginning with the case where \sigma is the initial permutation in the algorithm. As \pi is a cyclic signed permutation, \pi(1)<0 can only occur if \pi(1)<-1. Then \pi(1) cannot be a left-to-right maximum, so by the construction of \sigma using left-to-right maxima, we conclude \sigma(1)=\pi(1)<-1.
For our induction step, assume that \sigma(1)<-1, and let \sigma' be the signed permutation after swaps in the while loop in line ?. If no swaps involving \pm 1 or the element following it occur, then \sigma'(1)=\sigma(1)<-1. Hence, we consider the remaining cases where swaps do occur.
We first prove by contradiction that swaps cannot begin with \pm 1. If this occurs, \pm 1 must be swapped with \pm 2. As \sigma(1)<-1, order property ? implies that \pm 1 cannot be the last element in the cycle containing it. Then order property ? implies that \pm 2 appears as the last element of some cycle to the left of \pm 1 and \sigma(2)<\sigma(1)<-1. As \sigma(1)\notin \{-1,-2\}, we find that \sigma(1)\leq -3 and \sigma(2)<-3. However, \pm 2 is the last element in the cycle containing it, so \sigma(2)<-2 is the first element of that cycle, and this contradicts order property ?.
It now suffices to consider when the element following \pm 1 is swapped first. The sign of the element following \pm 1 does not change, so if swaps terminate after this, then \sigma'(1)<-1. Otherwise, \pm 1 must then be swapped with \pm 2. By Lemma 3.12, this can only occur if the elements following \pm 1 and \pm 2 in \sigma share the same sign, and hence must both be negative. We see that even in the case when \pm 1 and \pm 2 are swapped, we have \sigma'(1)<-1.
If 0\in \operatorname{Des}(\pi), then 0\in \operatorname{Des}(\phi(\pi)). Furthermore, in this case, we have that \pi(1)<-1 and \phi(\pi)(1)<-1.
If 0\notin \operatorname{Des}(\pi), then either 0\notin \operatorname{Des}(\sigma) or \sigma(1)=-1.
We will prove this using induction, again starting with the case where {\sigma} is the signed permutation at the start of the algorithm. If \pm 1 is not the last element of a cycle in {\sigma}, then {\sigma}(1)=\pi(1), so 0\notin \operatorname{Des}(\pi) implies 0\notin \operatorname{Des}(\sigma). Otherwise, {\sigma}(1) is the first element of the cycle containing \pm 1, and order property ? implies that \sigma(1)\geq -1, so either \sigma(1)=-1 or \sigma(1)\geq 1. The latter case is equivalent to 0\notin \operatorname{Des}(\sigma).
Now assume that the result holds for \sigma, and let \sigma' be the permutation obtained after swaps in the while loop in line ?. As in Lemma 3.28, we can assume that some swap involves \pm 1 or the element following it in the cycle notation. Our argument will require cases based on which property from the induction hypothesis holds for \sigma.
First, consider the induction hypotheses of 0\notin \operatorname{Des}(\sigma). We have two cases based on whether or not \pm 1 is involved in the first swap. If it is not, then the element following \pm 1, denoted x, is swapped first. As 0\notin \operatorname{Des}(\sigma), we know x>0. If swaps terminate after x is swapped, then the sign of the element following \pm 1 has not changed, so \sigma'(1)>0. Otherwise, \pm 1 is then swapped with \pm 2, and the result is \sigma'(1)=x>0. We conclude that 0\notin \operatorname{Des}(\sigma') holds in each case.
Alternatively, suppose \pm 1 is involved in the first swap. Then this occurs due to 1\in \operatorname{Des}_{\Delta}(\pi,\sigma), and order property ? for \sigma implies that either \pm 1 or \pm 2 is the last element of the cycle containing it. If \pm 1 is the last element of a cycle, then order property ? also implies \pm 2 appears in a cycle to the right of \pm 1 and \sigma(2)>\sigma(1)>0. Then the swap between \pm 1 and \pm 2 results in a positive number following 1 in the cycle notation. By Lemma 3.16, subsequent swaps cannot change this value, so \sigma'(1)=\sigma(2)>0 and 0\notin \operatorname{Des}(\sigma'). If \pm 2 is the last element of a cycle, then order property ? implies \sigma(2)\geq -2. Combined with the fact that \pm 1 is not in the same cycle as \pm 2, we conclude that either (-2) is a cycle in \sigma or \sigma(2)\geq 2. In the first case, (-1) will be a cycle in \sigma', and in the second case, \sigma'(1)\geq 1. These respectively correspond to \sigma'(1)=-1 or 0\notin \operatorname{Des}(\sigma').
Finally, consider the induction hypothesis that (-1) is a cycle in \sigma. In this case, a swap involving -1 can only occur if it is swapped with \pm 2 in the first and only swap performed on \sigma. Order property ? implies that \pm 2 appears as a non-last element in a cycle to the right of (-1), and \sigma(1)=-1<\sigma(2). In particular, we see that \sigma(2)>0, and the single swap in this while loop will then result in \sigma'(1)=\sigma(2)>0. Hence 0\notin \operatorname{Des}(\sigma'). By induction, we conclude that the result holds for all \sigma encountered in the algorithm.
We have that 0\in \operatorname{Des}_{\Delta}(\pi,\phi(\pi)) if and only if \pi(1)>0 and \phi(\pi)(1)=-1.
It is clear that if \pi(1)>0 and \phi(\pi)(1)=-1, then 0\in \operatorname{Des}_{\Delta}(\pi,\phi(\pi)), so it suffices to show the converse. Corollary 3.29 shows that 0\in \operatorname{Des}_{\Delta}(\pi,\phi(\pi)) can only occur if 0\notin \operatorname{Des}(\pi) and 0\in \operatorname{Des}(\phi(\pi)). Equivalently, \pi(1)>0 but \phi(\pi)(1)<0. By Lemma 3.30, this can only occur if \pi(1)>0 and \phi(\pi)(1)=-1.
For any \pi\in \operatorname{\mathfrak{C}}_{B,n+1}, define -\pi=[-1,-2,\ldots,-n] \pi, which is the permutation obtained by changing the signs of all elements in the one-line or cycle notation. Define the function \Phi:\operatorname{\mathfrak{C}}_{B,n+1}\to B_n by
where (-1)=[-1,2,\ldots,n]\in B_{n} is the signed permutation that interchanges \pm 1 with \mp 1 and fixes all other elements.
Recall the signed permutation
from Example 3.4. In that example, we found that
As 0\in \operatorname{Des}_{\Delta}(\pi,\phi(\pi)), we have that
Consider the signed permutation
which has descents \{1,4,5,7\}. For calculating \Phi(\pi), we first must apply Algorithm ? to
which has descents \{0,2,3,6,8\}. The algorithm will start with
which has descents \{0,2,3,6,7\}. The first iteration of the for loop will swap 8 and 7, followed by 4 and 5, resulting in
with descent set \{0,2,3,6\}. The algorithm terminates here, resulting in
with descent set \{1,4,5,7\}. As 0\notin \operatorname{Des}_{\Delta}(\pi,-\phi(-\pi)), we find that \Phi(\pi) is the permutation -\phi(-\pi) given above.
Let \pi\in \operatorname{\mathfrak{C}}_{B,n+1}. Then
We first consider \pi\in \operatorname{\mathfrak{C}}_{B,n+1}^+. By Corollary 3.26, we have
When equality holds, Corollary 3.31 implies that \phi(\pi)(1)=-1, so \operatorname{Des}(\pi)\cap \{0,1,\ldots,n-1\}=\operatorname{Des}((-1)\phi(\pi)). The function \Phi makes the appropriate choice so that \operatorname{Des}(\pi)\cap \{1,2,\ldots,n-1\}=\operatorname{Des}(\Phi(\pi)), so the result holds for any \pi\in \operatorname{\mathfrak{C}}_{B,n+1}^+.
Next, consider \pi\in \operatorname{\mathfrak{C}}_{B,n+1}^-. Then -\pi\in \operatorname{\mathfrak{C}}_{B,n+1}^+. From argument above, \operatorname{Des}(-\pi)\cap \{0,1,\ldots,n-1\} is equal to either \operatorname{Des}(\phi(-\pi)) or \operatorname{Des}((-1)\phi(-\pi)). This implies that \operatorname{Des}(\pi)\cap \{0,1,\ldots,n-1\} is equal to either \operatorname{Des}(-\phi(-\pi)) or \operatorname{Des}(-(-1)\phi(-\pi)), and \Phi makes the appropriate choice so that \operatorname{Des}(\pi)\cap \{0,1,\ldots,n-1\}=\operatorname{Des}(\Phi(\pi)).
Define \psi:B_n\to \operatorname{\mathfrak{C}}_{B,n+1}^+ to be the function mapping each \sigma\in B_n to its output from Algorithm ?.
The initial permutation in Algorithm ? applied to \sigma is \pi^{(m)}.
Let 0\leq i\leq h and x\in [n-1]. If x\in \operatorname{Des}_{\Delta}(\pi^{(j,i)},\sigma^{(j,i)}), then one of the following must be true:
\pm x appears as the last element in some cycle of \sigma^{(j,i)}, \pm (x+1) appears as a non-last element in some cycle to the right of \pm x, and \pi^{(j,i)}(x)>\pi^{(j,i)}(x+1)=\sigma^{(j,i)}(x+1)>\sigma^{(j,i)}(x), or
\pm (x+1) appears as the last element in some cycle of \sigma^{(j,i)}, \pm x appears as a non-last element in some cycle to the right of \pm (x+1), and \pi^{(j,i)}(x+1)>\pi^{(j,i)}(x)=\sigma^{(j,i)}(x)>\sigma^{(j,i)}(x+1).
Let z be the last element of \sigma_j^{(j,0)} and \epsilon\in \{-1,1\}. If P_{\pi,\sigma^{(j,0)}}(|z|,|z+\epsilon|) is False, then P_{\pi^{(j,0)},\sigma}(|z|,|z+\epsilon|) is False.
We use proof by contradiction. Assume that P_{\pi,\sigma^{(j,0)}}(|z|,|z+\epsilon|) is False and P_{\pi^{(j,0)},\sigma}(|z|,|z+\epsilon|) is True. Corollary 3.26 states that \operatorname{Des}_{\Delta}(\pi,\sigma)\cap [n-1]=\emptyset, so this combination can only occur if P_{\pi^{(j,0)},\sigma^{(j,0)}}(|z|,|z+\epsilon|) is True. Consequently, we have that \min\{|z|,|z+\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi^{(j,0)},\sigma^{(j,0)}), and Lemma 4.3 implies that
Swap property ? on all signed permutations prior to \sigma^{(j,0)} in Algorithm ? implies that \pm z has been the last element of cycle j since the beginning of Algorithm ?, so \pi(|z|)=\sigma^{(1)}_{j+1,1} . However, order property ? for \sigma with (Proof) then implies that \pi(|z|)=\sigma^{(1)}_{j+1,1}>\pi(|z+\epsilon|). This contradicts P_{\pi,\sigma^{(j,0)}}(|z|,|z+\epsilon|) being False and \sigma^{(j,0)}(|z+\epsilon|)>\sigma^{(j,0)}(|z|) from (Proof).
Suppose \pm z is the last element in cycle j of \sigma^{(j,0)} and is swapped with \pm (z+\epsilon) in Algorithm ?, so \pm (z+i\epsilon) is the last element in cycle j of \sigma^{(j,i)} for 0\leq i\leq h. Then P_{\pi^{(j,i)},\sigma}(|z+i\epsilon|,|z+(i-1)\epsilon|) is True for all 1\leq i\leq h.
Order property ? for the respective last elements \pm z,\pm (z+\epsilon),\ldots,\pm (z+h\epsilon) of \sigma^{(j,0)},\sigma^{(j,1)},\ldots,\sigma^{(j,h)} implies that
and Corollary 3.26 implies the corresponding order
It suffices now to show that \pi^{(j,i)}(|z+i\epsilon|)>\pi^{(j,i)}(|z+(i-1)\epsilon|).
We turn our attention to \sigma^{(j,i-1)}, where order property ? implies that the last element \pm (z+(i-1)\epsilon)\in \sigma_j^{(j,i-1)} was swapped with \pm (z+i\epsilon), which appears as a non-last element in some cycle to its right. After swaps occur, \pm (z+(i-1)\epsilon) is not the last element of a cycle in \sigma^{(j,i)}, and from this, we conclude
Swap property ? for \sigma^{(j,i-1)} implies that \sigma^{(j,i-1)}(|z+i\epsilon|)<\sigma_{j+1,1}^{(j,i-1)}, so after swaps are performed, \sigma^{(j,i)}(|z+(i-1)\epsilon|)<\sigma_{j+1,1}^{(j,i)}. Combined, we see that
Comparing with (Proof), we conclude P_{\pi^{(j,i)},\sigma}(|z+i\epsilon|,|z+(i-1)\epsilon|) is True.
Suppose \pm z is the last element in cycle j of \sigma^{(j,0)} and is swapped with \pm (z+\epsilon) in Algorithm ?, so \pm (z+i\epsilon) is the last element in cycle j of \sigma^{(j,i)} for 0\leq i\leq h. If P_{\pi^{(j,h)},\sigma}(|z+h\epsilon|,|z+(h+1)\epsilon|) is True, then \pi^{(j,h)}(|z+(h-1)\epsilon)<\pi^{(j,h)}(|z+(h+1)\epsilon|).
Since Algorithm ? terminated iteration j of the for loop at \sigma^{(j,h)}, we know that P_{\pi,\sigma^{(j,h)}}(|z+h\epsilon|,|z+(h+1)\epsilon|) is False. Corollary 3.26 implies \operatorname{Des}(\pi)\cap [n-1]=\operatorname{Des}(\sigma), and combined with P_{\pi^{(j,h)},\sigma}(|z+h\epsilon|,|z+(h+1)\epsilon|) being True, we conclude that \min\{|z+h\epsilon|,|z+(h+1)\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi^{(j,h)},\sigma^{(j,h)}). As \pm (z+h\epsilon) is the last element of \sigma_j^{(j,h)}, Lemma 4.3 implies that \pm (z+(h+1)\epsilon) must appear as the non-last element in some cycle to the right of \sigma_j^{(j,h)} and
Since P_{\pi,\sigma^{(j,h)}}(|z+h\epsilon|,|z+(h+1)\epsilon|) is False, we conclude
We now turn our attention to \sigma^{(j,h-1)}. As \pm (z+(h+1)\epsilon) is not the last element in a cycle of \sigma_j^{(j,h)}, swap properties ? and ? for \sigma^{(j,h-1)} imply that \pm (z+(h+1)\epsilon) cannot be the last element in a cycle of \sigma^{(j,h-1)}. Additionally, order property ? for \sigma^{(j,h-1)} implies that \pm (z+h\epsilon) was not the last element in a cycle of \sigma^{(j,h-1)}. Since \pm (z+(h+1)\epsilon) and \pm (z+h\epsilon) are non-last elements in the cycles of \sigma^{(j,h-1)} containing them, order property ? for \sigma^{(j,h-1)} implies \min\{|z+h\epsilon|,|z+(h+1)\epsilon|\}\notin \operatorname{Des}_{\Delta}(\pi,\sigma^{(j,h-1)}), so combined with (Proof), we see that \sigma^{(j,h-1)}(|z+(h+1)\epsilon|)>\sigma^{(j,h-1)}(|z+h\epsilon|). Then after performing swaps to obtain \sigma^{(j,h)}, we have that
Additionally, these swaps result in \pm (z+(h-1)\epsilon) being a non-last element of a cycle in \sigma^{(j,h)}. As both \pm (z+(h-1)\epsilon) and \pm (z+(h+1)\epsilon) are non-last elements in the cycles containing them, we conclude that (?) is equivalent to \pi^{(j,h)}(|z+(h+1)\epsilon|)>\pi^{(j,h)}(|z+(h-1)\epsilon|).
The functions \phi:\operatorname{\mathfrak{C}}_{B,n+1}^+\to B_n and \psi:B_n\to \operatorname{\mathfrak{C}}_{B,n+1}^+ are inverses.
It suffices to show that \psi(\phi((\pi))=\pi for every \pi\in \operatorname{\mathfrak{C}}_{B,n+1}^+. As \operatorname{\mathfrak{C}}_{B,n+1}^+ and B_n are finite sets of the same size 2^n\cdot n!, it will follow that \phi and \psi are bijections that must be inverses of one another. By Lemma 4.2, we know that Algorithm ? applied to \sigma=\phi(\pi) will begin with \pi^{(m)}. It suffices to show that if \pi^{(j+1)} is the permutation at the start of iteration j of the for loop in Algorithm ?, then \pi^{(j)} will be the result at the end, as the result will then follow by induction on j=m-1,m-2,\ldots,1.
We first consider the case where no swaps are performed in iteration j of the for loop in Algorithm ?, so \sigma^{(j)}=\sigma^{(j+1)} and \pi^{(j)}=\pi^{(j+1)}. This occurs since P_{\pi,\sigma^{(j)}}(|z|,|z+\epsilon|) is False for each value of \epsilon\in \{-1,1\}. Then Lemma 4.4 implies that P_{\pi^{(j+1)},\sigma}(|z|,|z+\epsilon|) is False for each value of \epsilon\in \{-1,1\}, so iteration j of the for loop in Algorithm ? will also perform no swaps. Hence, it will begin and terminate with \pi^{(j+1)}=\pi^{(j)}.
Alternatively, suppose swaps do occur. As we have throughout this section, we will use \sigma^{(j,0)},\sigma^{(j,1)},\ldots,\sigma^{(j,h)} for the signed permutations in iteration j of the for loop of Algorithm ?, and the respective last elements of cycle j will be \pm z,\pm (z+\epsilon),\ldots,\pm (z+h\epsilon). By Lemma 4.5, Lemma 4.6, the for loop of Algorithm ? will start with \pi^{(j,h)}, swap \pm (z+h\epsilon) and \pm (z+(h-1)\epsilon), and then subsequently swap the elements preceding the ones swapped whenever another descent is affected by the prior swap. This reverses exactly the swaps of the while loop in line ? of Algorithm ?, and hence results in \pi^{(j,h-1)}. Continuing with Lemma 4.5 and a similar argument, we conclude that this iteration of the for loop in Algorithm ? will then produce \pi^{(j,h-2)},\ldots,\pi^{(j,0)}. It suffices now to show that the loop will terminate at \pi^{(j,0)} due to P_{\pi^{(j,0)},\sigma}(|z|,|z-\epsilon|) being False. If P_{\pi,\sigma^{(j,0)}}(|z|,|z-\epsilon|) is False, then this follows from Lemma 4.4, so it suffices to consider when P_{\pi,\sigma^{(j,0)}}(|z|,|z-\epsilon|) is True.
In this case, order property ? for \sigma^{(j,0)} implies that both \pm (z+\epsilon) and \pm (z-\epsilon) appear as non-last elements in the cycles of \sigma^{(j,0)} containing them. Additionally, Algorithm ? chose to swap \pm z with \pm (z+\epsilon) when both P_{\pi,\sigma^{(j,0)}}(|z|,|z-\epsilon|) and P_{\pi,\sigma^{(j,0)}}(|z|,|z+\epsilon|) were True, so \sigma^{(j,0)}(|z+\epsilon|)>\sigma^{(j,0)}(|z-\epsilon|). Combined with order property ? for \sigma^{(j,0)}, it must be that
while \pi(|z|)>\pi(|z+\epsilon|) and \pi(|z|)>\pi(|z-\epsilon|). Corollary 3.26 then implies \sigma(|z|)>\sigma(|z+\epsilon|) and \sigma(|z|)>\sigma(|z-\epsilon|). The swaps producing \pi^{(j,0)} from \pi^{(j,1)} removed the descent \min\{|z|,|z+\epsilon|\}\in \operatorname{Des}_{\Delta}(\pi^{(j,1)},\sigma), so \pi^{(j,0)}(|z|)>\pi^{(j,0)}(|z+\epsilon|). Combined with (?), we conclude that \pi^{(j,0)}(|z|)>\pi^{(j,0)}(|z-\epsilon|). As this matches the order of \sigma(|z|)>\sigma(|z-\epsilon|), we see that P_{\pi^{(j,0)},\sigma}(|z|,|z-\epsilon|) is False, so iteration j of the for loop will terminate at \pi^{(j)}=\pi^{(j,0)}.
For any \sigma\in B_n, the following hold.
We have \psi(\sigma),\psi((-1)\sigma)\in \operatorname{\mathfrak{C}}_{B,n+1}^+ and -\psi(-\sigma),-\psi(-(-1)\sigma)\in \operatorname{\mathfrak{C}}_{B,n+1}^-.
The parities of the \operatorname{neg} statistic on \sigma, \psi(\sigma), and -\psi(-(-1)\sigma) are the same, and the parities of the \operatorname{neg} statistic on (-1)\sigma, \psi((-1)\sigma), and -\psi(-\sigma) are the same.
We have \Phi^{-1}(\{\sigma,(-1)\sigma\})=\{\psi(\sigma),\psi((-1)\sigma),-\psi(-\sigma),-\psi(-(-1)\sigma)\}. Furthermore, these four elements belong to different subsets \operatorname{\mathfrak{C}}_{B,n+1}^+\cap \operatorname{\mathfrak{C}}_{D,n+1}, \operatorname{\mathfrak{C}}_{B,n+1}^+\cap \overline{\operatorname{\mathfrak{C}}_{D,n+1}}, \operatorname{\mathfrak{C}}_{B,n+1}^-\cap \operatorname{\mathfrak{C}}_{D,n+1}, and \operatorname{\mathfrak{C}}_{B,n+1}^-\cap \overline{\operatorname{\mathfrak{C}}_{D,n+1}} of \operatorname{\mathfrak{C}}_{B,n+1}.
For (a), observe that in Algorithm ?, n+1 is inserted in the cycle notation and is not involved in any swaps, as a descent at position n is not considered in the algorithm. This immediately implies that \psi(\sigma),\psi((-1)\sigma),\psi(-\sigma),\psi(-(-1)\sigma)\in \operatorname{\mathfrak{C}}_{B,n+1}^+. Changing the signs of all elements after applying \psi to a signed permutation results in -\psi(-\sigma),-\psi(-(-1)\sigma)\in \operatorname{\mathfrak{C}}_{B,n+1}^-.
For (b), observe that inserting n+1 and performing the swaps in Algorithm ? will not change the number of negatives appearing in the cycle notation, so \operatorname{neg}(\sigma)=\operatorname{neg}(\psi(\sigma)). Turning our attention to -\psi(-(-1)\sigma), we use this to find
and since \operatorname{neg}((-1)\sigma)\in \{\operatorname{neg}(\sigma)-1,\operatorname{neg}(\sigma)+1\}, this has the same parity as \operatorname{neg}(\sigma). The corresponding result for (-1)\sigma, \psi((-1)\sigma), and -\psi(-\sigma) follows by replacing \sigma with (-1)\sigma throughout.
Finally, we consider (c). From Theorem 4.7, \phi and \psi are inverses. Using this with the definition of \Phi, any pre-images of \{\sigma,(-1)\sigma\} in \operatorname{\mathfrak{C}}_{B,n+1}^+ must be in \{\psi(\sigma),\psi((-1)\sigma)\}, and any pre-images in \operatorname{\mathfrak{C}}_{B,n+1}^- must be in \{-\psi(-\sigma),-\psi(-(-1)\sigma)\}. Applying \Phi to \psi(\sigma),\psi((-1)\sigma)\in \operatorname{\mathfrak{C}}_{B,n+1}^+ or -\psi(-\sigma),-\psi(-(-1)\sigma)\in \operatorname{\mathfrak{C}}_{B,n+1}^- must also result in \sigma or (-1)\sigma, so combined, we conclude
The remaining claims in (c) now follow from casework based on whether \operatorname{neg}(\sigma) is even or odd, combined with (a) and (b).
Let \sigma\in B_n, and suppose \sigma(1)=1. Define \pi_1=\psi(\sigma), \pi_2=\psi((-1)\sigma), \pi_3=-\psi(-\sigma), and \pi_4=-\psi(-(-1)\sigma). Then \Phi(\pi_1)=\Phi(\pi_2)=\sigma and \Phi(\pi_3)=\Phi(\pi_4)=(-1)\sigma.
We first consider \Phi(\pi_1), where \pi_1\in \operatorname{\mathfrak{C}}_{B,n+1}^+ by Lemma 4.8. If \Phi(\pi_1)=\phi(\pi_1), then Theorem 4.7 implies that \Phi(\pi_1)=\phi(\psi(\sigma))=\sigma, so the claim holds. Otherwise, \Phi(\pi_1)=(-1)\phi(\pi_1). By Corollary 3.31, this can only occur if 0\notin \operatorname{Des}(\pi_1) and \phi(\pi_1)(1)=-1. From Lemma 4.8 and the assumption \sigma(1)=1, we can conclude \phi(\pi_1)=(-1)\sigma, so \Phi(\pi_1)=(-1)(-1)\sigma=\sigma.
We next consider \Phi(\pi_2), where \pi_2\in \operatorname{\mathfrak{C}}_{B,n+1}^+ by Lemma 4.8. If \Phi(\pi_2)=(-1)\phi(\pi_2), then \Phi(\pi_2)=(-1)(-1)\sigma=\sigma, so the claim holds. Otherwise, \Phi(\pi_2)=\phi(\pi_2)=(-1)\sigma. As \sigma(1)=1, we know that (-1) is a cycle in (-1)\sigma and 0\in \operatorname{Des}((-1)\sigma). Additionally, \Phi(\pi_2)=\phi(\pi_2) occurs when 0\notin \operatorname{Des}_{\Delta}(\pi_2,\phi(\pi_2)), so 0\in \operatorname{Des}(\pi_2). However, 0\in \operatorname{Des}(\pi_2) and \phi(\pi_2)(1)=(-1)\sigma(1)=-1 contradicts Corollary 3.29.
We now consider \Phi(\pi_3), where \pi_3\in \operatorname{\mathfrak{C}}_{B,n+1}^- by Lemma 4.8. In the case \Phi(\pi_3)=-(-1)\phi(-\pi_3), we find
so the claim holds. Otherwise, \Phi(\pi_3)=-\phi(-\pi_3)=\sigma. As \sigma(1)=1, we know 0\notin \operatorname{Des}(\sigma), so Theorem 3.35 implies that 0\notin \operatorname{Des}(\pi_3) and 0\in \operatorname{Des}(-\pi_3). Additionally, \Phi(\pi_3)=-\phi(-\pi_3)=\sigma implies that \phi(-\pi_3)=-\sigma, and since \sigma(1)=1, we see that -\sigma(1)=-1. However, 0\in \operatorname{Des}(-\pi_3) and \phi(-\pi_3)(1)=-\sigma(1)=-1 similarly contradicts Corollary 3.29.
Finally, consider \Phi(\pi_4), where \pi_4\in \operatorname{\mathfrak{C}}_{B,n+1}^- by Lemma 4.8. If \Phi(\pi_4)=-\phi(-\pi_4), then we find
and the claim holds. Otherwise,
As \sigma(1)=1, we know that 0\notin \operatorname{Des}(\sigma), and Theorem 3.35 implies 0\notin \operatorname{Des}(\pi_4). We conclude 0\in \operatorname{Des}(-\pi_4) and 0\notin \operatorname{Des}(\phi(-\pi_4))=\operatorname{Des}(-(-1)\sigma), which again contradicts Corollary 3.29.
Let \sigma\in B_n, and suppose \sigma(1)=-1. Define \pi_1=\psi(\sigma), \pi_2=\psi((-1)\sigma), \pi_3=-\psi(-\sigma), and \pi_4=-\psi(-(-1)\sigma). Then \Phi(\pi_1)=\Phi(\pi_2)=(-1)\sigma and \Phi(\pi_3)=\Phi(\pi_4)=\sigma.
Define \Psi_{\operatorname{\mathfrak{C}}_{D,n+1}}:B_n\to \operatorname{\mathfrak{C}}_{D,n+1} by
Define \Psi_{\overline{\operatorname{\mathfrak{C}}_{D,n+1}}}:B_n\to \overline{\operatorname{\mathfrak{C}}_{D,n+1}} by
The functions \Phi_{\operatorname{\mathfrak{C}}_{D,n+1}} and \Psi_{\operatorname{\mathfrak{C}}_{D,n+1}} are inverses, and the functions \Phi_{\overline{\operatorname{\mathfrak{C}}_{D,n+1}}} and \Psi_{\overline{\operatorname{\mathfrak{C}}_{D,n+1}}} are inverses.
We will show this for \Phi_{\operatorname{\mathfrak{C}}_{D,n+1}} and \Psi_{\operatorname{\mathfrak{C}}_{D,n+1}}. The argument for \Phi_{\overline{\operatorname{\mathfrak{C}}_{D,n+1}}} and \Psi_{\overline{\operatorname{\mathfrak{C}}_{D,n+1}}} is similar, so we leave this to the reader. From Lemma 4.8, it is straightforward to show that the six cases of \Psi_{\operatorname{\mathfrak{C}}_{D,n+1}} produce elements in \operatorname{\mathfrak{C}}_{D,n+1}, so \Psi_{\operatorname{\mathfrak{C}}_{D,n+1}} is well-defined. It suffices now to show \Phi_{\operatorname{\mathfrak{C}}_{D,n+1}}(\Psi_{\operatorname{\mathfrak{C}}_{D,n+1}}(\sigma))=\sigma for all \sigma\in B_n. As \operatorname{\mathfrak{C}}_{D,n+1} and B_n are finite sets of the same size 2^n\cdot n!, it will follow that \Phi_{\operatorname{\mathfrak{C}}_{D,n+1}} and \Psi_{\operatorname{\mathfrak{C}}_{D,n+1}} are bijections that must be inverses of one another.
We first consider when \sigma(1)\notin \{-1,1\}, where we will use the fact that \phi and \psi are inverses from Theorem 4.7. In the case where \operatorname{neg}(\sigma) is even, we have that \Psi_{\operatorname{\mathfrak{C}}_{D,n+1}}(\sigma)=\psi(\sigma)\in \operatorname{\mathfrak{C}}_{B,n+1}^+ and \phi(\Psi_{\operatorname{\mathfrak{C}}_{D,n+1}}(\sigma))=\sigma. Since \sigma(1)\notin \{-1,1\}, Lemma 3.30 implies that \Phi will not multiply by (-1), so \Phi_{\operatorname{\mathfrak{C}}_{D,n+1}}(\Psi_{\operatorname{\mathfrak{C}}_{D,n+1}}(\sigma))=\phi(\Psi_{\operatorname{\mathfrak{C}}_{D,n+1}}(\sigma))=\sigma. In the case where \operatorname{neg}(\sigma) is odd, we have that \Psi_{\operatorname{\mathfrak{C}}_{D,n+1}}=-\psi(-\sigma)\in \operatorname{\mathfrak{C}}_{B,n+1}^- and -\phi(-\Psi_{\operatorname{\mathfrak{C}}_{D,n+1}}(\sigma))=\sigma. Lemma 3.30 similarly implies that \Phi_{\operatorname{\mathfrak{C}}_{D,n+1}} will not multiply by (-1), so \Phi_{\operatorname{\mathfrak{C}}_{D,n+1}}(\Psi_{\operatorname{\mathfrak{C}}_{D,n+1}}(\sigma))=\sigma.
We now consider when \sigma(1)\in \{-1,1\}. If \sigma(1)=1 and \operatorname{neg}(\sigma) is even, then Lemma 4.9 implies
If \sigma(1)=1 and \operatorname{neg}(\sigma) is odd, then Lemma 4.9 implies
If \sigma(1)=-1 and \operatorname{neg}(\sigma) is even, then Corollary 4.10 implies
Finally, if \sigma(1)=-1 and \operatorname{neg}(\sigma) is odd, then Corollary 4.10 implies
Let (X_n)_{n\geq 1} be the random variables corresponding to the descent statistic on (\operatorname{\mathfrak{C}}_{B,n})_{n\geq 1}, (\operatorname{\mathfrak{C}}_{D,n})_{n\geq 1}, or (\overline{\operatorname{\mathfrak{C}}_{D,n}})_{n\geq 1} with means (\mu_n)_{n\geq 1} and variances (\sigma_n^2)_{n\geq 1}. The standardized random variable (X_n-\mu_n)/\sigma_n converges in distribution to the standard normal distribution.
Fix (\Omega_n)_{n\geq 1} to be either (\operatorname{\mathfrak{C}}_{B,n})_{n\geq 1}, (\operatorname{\mathfrak{C}}_{D,n})_{n\geq 1}, or (\overline{\operatorname{\mathfrak{C}}_{D,n}})_{n\geq 1}. Define X_n:\Omega_n\to \mathbb{R} to be the random variable corresponding to the descent statistic on \pi\in \Omega_n generated uniformly at random, and define X_n':\Omega_n\to \mathbb{R} to be the corresponding descent statistic on \Phi(\pi), where \Phi is the function from Theorem 1.2. Using these, express
Theorem 1.2 implies that X_n-X'_n is a random variable that takes values in \{0,1\}. When n is sufficiently large, Lemma 2.11 and Theorem 2.8 imply \mu_n=n/2 and \sigma_n=\sqrt{(n+1)/12}. Combining these observations, it is straightforward to verify (?) to show that
Corollary 1.3 implies that the distribution of X'_n coincides with the distribution of the descent statistic for elements of B_{n-1} generated uniformly at random. Hence, Theorem 2.8 implies that {(X'_n-\mu_{n-1})}/{\sigma_{n-1}} \xrightarrow{d} \mathcal{N}(0,1). Returning to (Proof), we now have that
so the result follows from Theorem 2.7.
Let (X_n)_{n\geq 1} be the random variables corresponding to the flag major index statistic on (\operatorname{\mathfrak{C}}_{B,n})_{n\geq 1}, (\operatorname{\mathfrak{C}}_{D,n})_{n\geq 1}, or (\overline{\operatorname{\mathfrak{C}}_{D,n}})_{n\geq 1} with means (\mu_n)_{n\geq 1} and variances (\sigma_n^2)_{n\geq 1}. The standardized random variable (X_n-\mu_n)/\sigma_n converges in distribution to the standard normal distribution.
We proceed as in the proof of Theorem 5.1 with the flag major index statistic for X_n and X_n' to obtain (Proof) again. Theorem 1.2, Lemma 2.12, and Theorem 2.9 imply that when n us sufficiently large, we have 0\leq X_n-X_n'\leq 2n+1, \mu_n=n^2/4, and \sigma_n=\sqrt{4n^3+6n^2-n}/6. From this, we find that
One can similarly show that (Proof) holds, and the result again follows from Theorem 2.7.
The absolute values and negative signs in Algorithm ? can be omitted. Furthermore, 0\notin \operatorname{Des}(\pi) and 0\notin \operatorname{Des}(\phi(\pi)), so \Phi(\pi)=\phi(\pi).
In lines ? and ? of Algorithm ?, P_{\pi,\sigma}(z,z+\epsilon) is True precisely when \pi(z)>\pi(z+\epsilon) and \sigma(z)<\sigma(z+\epsilon).
The while loop on line ? terminates when the first element of \sigma_j is swapped or the elements preceding the ones swapped differ by more than 1.
For any positive integers n and r, there exists a function \Phi: \operatorname{\mathfrak{C}}_{S,n+1,r}\to S_{n,r} with the following properties:
for all (\omega,\tau)\in \operatorname{\mathfrak{C}}_{S,n+1,r}, we have \operatorname{Des}(\omega,\tau)\cap \{1,2,\ldots,n-1\}=\operatorname{Des}(\Phi(\omega,\tau)), and
the restriction of \Phi to the elements of \operatorname{\mathfrak{C}}_{S,n+1,r} with any fixed color is a bijection.
Let \phi:\operatorname{\mathfrak{C}}_{S,n+1}\to S_n be the bijection from Theorem 1.1. Extend this to a map \Phi on \operatorname{\mathfrak{C}}_{S,n+1,r} by defining \Phi(\omega,\tau)=(\phi(\omega),\tau|_{[n]}), where \tau|_{[n]} denotes the restriction of \tau. For any i\in [n-1], we consider a descent at position i by focusing on the one-line notations for (\omega,\tau) and \Phi(\omega,\tau).
If \tau(i)\neq \tau(i+1), then a descent at position i in (\omega,\tau) and \Phi(\omega,\tau) is determined by the relative order of \tau(i)=\tau|_{[n]}(i) and \tau(i+1)=\tau|_{[n]}(i+1) as elements in \mathbb{Z}. Hence, i\in \operatorname{Des}(\omega,\tau) if and only if i\in \operatorname{Des}(\Phi(\omega,\tau)).
If \tau(i)=\tau(i+1), then a descent at position i in (\omega,\tau) and \Phi(\omega,\tau) is determined by whether or not there is a descent at position i in \omega and \phi(\omega). By Theorem 1.1, i\in \operatorname{Des}(\omega) if and only if i\in \operatorname{Des}(\phi(\omega)), so we conclude i\in \operatorname{Des}(\omega,\tau) if and only if i\in \operatorname{Des}(\Phi(\omega,\tau)).
Combined, we see that (a) holds.
For (b), we fix a color c\in \mathbb{Z}. We map each (\omega',\tau')\in S_{n,r} to an element (\omega,\tau)\in \operatorname{\mathfrak{C}}_{S,n+1,r} as follows:
define \omega=\phi^{-1}(\omega'), and
extend \tau':[n]\to \mathbb{Z}_r to \tau:[n+1]\to \mathbb{Z}_r by defining
It is straightforward to show that this is the inverse of \Phi on the elements of \operatorname{\mathfrak{C}}_{S,n+1,r} with color c.
In general, the bijection in the proof of Theorem B.1 does not preserve the descent at position n. This descent for (\omega,\tau)\in \operatorname{\mathfrak{C}}_{S,n+1,r} is determined by the relative order of (\omega,\tau)(n) and (\omega,\tau)(n+1), while this only depends on \tau(n) for \Phi(\omega,\tau)\in S_{n,r}. Additionally, n+1 is a possible descent for (\omega,\tau)\in \operatorname{\mathfrak{C}}_{S,n+1,r}, while it is not a possible descent for \Phi(\omega,\tau)\in S_{n,r}. Combined, we see that \operatorname{Des}(\omega,\tau) and \operatorname{Des}(\Phi(\omega,\tau)) can potentially differ at two values, n and n+1.
Let (X_n)_{n\geq 1} be the random variables corresponding to either the descent or flag major index statistic on the cyclic permutations in S_{n,r} with colors in any fixed, nonempty subset of \mathbb{Z}_r. Letting (\mu_n)_{n\geq 1} and (\sigma_n^2)_{n\geq 1} denote the respective means and variances, the standardized random variable (X_n-\mu_n)/\sigma_n converges in distribution to the standard normal distribution.
The text on this page is from arXiv:2508.02432v1 by Kevin Liu, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.
The record of each item is in analysis/objects/2508/2508.02432.jsonl.