# In [[mathematics]] the Robinson-Schensted-[[Knuth]] correspondence is a [[biject ion]] between generalized permutations (as 2 lined arrays) and pairs

of [[Young_ tableau#Tableaux semi-standard Young Tableaux]]. [[Richard_P._Stanley Richard St anley]] uses this term for these tableaux and abbreviates it as SSYT <ref name=S tanley> Stanley, Richard P., Enumerative Combinatorics, Volume2. Cambridge Unive rsity Press. ISBN 0-521-55309-1, 0-521-56069-1. Page 316-380</ref>. Donald Knuth uses the term generalized Young Tableaux <ref name=Knuth> Knuth, Donald E., Per mutations, matrices, and generalized Young tableaux. Pacific J. Math. Volume 34, Number 3 (1970), 709-727. Available at http://projecteuclid.org/DPubS?verb=Disp lay&version=1.0&service=UI&handle=euclid.pjm/1102971948&page=record</ref>. == Introduction == The [[Robinson-Schensted algorithm]] helps establish a [[bijective]] mapping bet ween [[permutations]] and pairs of [[Young_tableau Young Tableaux]], both having shape $\lambda$: $\pi \longleftrightarrow (\lambda,P,Q)$ where $\pi \in S_n$ is a permutation of order $n$ and <ma th>P,Q[/itex] are Young Tableau of shape $\lambda$. ===Generalized permutations=== A ''generalized permutation'' or two-line array $w$ is defines by<ref name=Stanley/>: $p = \begin{pmatrix}i_1 & i_2 & \ldots & i_m\j_1 & j_2 & \ldots & j_m\end {pmatrix}$ where, # $i_1 \leq i_2 \leq i_3 \ldots \leq i_m$ # if $i_r = i_s$ and $r \leq s$ then $j_r \leq j_s</m ath> '''Example''': [itex]p = \begin{pmatrix}1 & 1 & 1 & 2 & 2 & 3 & 3\ 1 & 3 & 3 & 2 & 2 & 1 & 2\end{pmatrix}$ By extending the Robinson-Schensted algorithm to generalized permutations we can obtain [[one-to-one]] mappings from these types of permutations to ordered pair s, $(P,Q)$, where $P$ and $Q$ are SSYT of the s ame shape. ==The Robinson-Schensted-Knuth correspondence== The Robinson-Schensted-Knuth (RSK) algorithm works almost exactly like the [[Rob inson-Schensted algorithm]], the only difference being that RSK takes a generali zed permutation as input. Stanley <ref name=Stanley/> uses this moniker. The ba sic operation consists of an ''insertion operation'' defined as $P \longlef tarrow k$ of a positive integer $k$ into a SSYT $P$. Let $A = (a_{ij})_{i,j \geq 1}$ be a matrix with non-negative element s of dimension $m \times n$. Let $p_A$ be a generalized pe rmutation associated with $A$, defined as: $p_A = \begin{pmatrix}i_1 & i_2 & \ldots & i_m\j_1 & j_2 & \ldots & j_m\en d{pmatrix}$ where in addition to the 2 rules specified in the definition of a generalized pe rmutaion $p_A$ needs to satisfy: # For each $(i,j)$, there must be $a_{ij}$ values of <math >r[/itex] for which $(i_r,j_r) = (i,j)$. It is easy to see that there is a bijective mapping from $A$ to <math >p_A[/itex]. '''Example''': If $p_A = \begin{pmatrix}1 & 1 & 1 & 2 & 2 & 3 & 3\ 1 & 3 & 3 & 2 & 2 & 1 & 2\end{pmatrix}$ then $A = \begin{bmatrix}1 & 0 & 2 \ 0 & 2 & 0 \ 1 & 1 & 0 \end{bmatrix}$ If we apply the RSK algorithm on the permutation $p_A$ we get the fol lowing theorem called the Robinson-Schensted-Knuth correspondence theorem<ref na me=Knuth/>. '''Theorem 1''': There is a one-one correspondence between matrix $A = (a_{ ij})_{i,j \geq 1}$ (and by implication permutation $p_A$) and t he ordered pairs $(P,Q)$, where $P$ and $Q$ hav e the same shape. In addition integer $i$ occurs exactly $a_{i1} + a_{i2} + \ldots + a_{in}$ times in $Q$ and the integer <math >j[/itex] occurs exactly $a_{1j} + a_{2j} + \ldots + a_{mj}$ times in $P$. == Combinatorial properties of RSK correspondence == ===RSK and permutation matrices=== If $A$ is a [[permutation matrix]] then RSK outputs standard Young Ta bleaux, $P,Q$ of the same shape $\lambda$. Conversely, if $P,Q$ are SYT having the same shape $\lambda$, then the co rresponding matrix $A$ is a permutation matrix. As a result of this p roperty by simply comparing the cardinalities of the two sets on the two sides o f the bijective mapping we get the following corollary: '''Corollary 1''': For each $n \geq 1$ we have $\sum_{\lambda\vd ash n} (f^\lambda)^2= n!$ where $\lambda\vdash n$ means $\lambda$ varies over all [[ Partition (number theory) partition]]s of $n$ and $f^\lambda</ma th> is the number of standard Young tableaux of shape [itex]\lambda$. By examining the structure of the Robinson-Schensted-K algorithm we can prove th e following theorem:<ref name=AOP> Kunth, Donald E., The Art of Programming, Vol ume 3/ Sorting and Searching. Addison-Wesley, 1973. Page 54-58</ref> '''Theorem 2''': If the permutation $\sigma$ corresponds to a triple $(\lambda,P,Q)$, then the [[Permutation#Product_and_inverse inverse p ermutation]], $\sigma^{-1}$, corresponds to $(\lambda,Q,P)</math >. This leads to the following surprising corollary that links the number of involu tions on [itex]S_n$ with the number of tableaux that can be formed from <m ath>S_n[/itex] (An ''involution'' is a permutation that is its own [[Permutation #Product_and_inverse inverse]])<ref name=AOP/>: '''Corollary 2''': The number of tableaux that can be formed from $\{1,2,3 , \ldots,n\}$ is equal to the number of involutions on $\{1,2,3, \ldo ts,n\}$. ''Proof'': If $\pi$ is an involution corresponding to $(P,Q)</ma th>, then [itex]\pi = \pi^-$ corresponds to $(Q,P)$; hence <mat h>P = Q[/itex]. Conversely, if $\pi$ is any permutation corresponding to $(P,P)$, then $\pi^-$ also corresponds to $(P,P)< /math>; hence [itex]\pi = \pi^-$. So there is a one-one correspondence be tween involutions $\pi$ and tableax $P$ The number of involutions on $\{1,2,3, \ldots,n\}$ is given by the re currence: $a(n) = a(n-1)+(n-1)a(n-2)$ Where $a(1) = 1,a(2) = 2$. By solving this recurrence we can get the number of involutions on $\{1,2,3, \ldots,n\}$, $I(n) = n!\sum_{k=0}^{\lfloor n/2 \rfloor} \frac{1}{2^kk!(n-2k)!}$ ===Symmetry of RSK<ref name=Stanley/>=== Let $A$ be a matrix with non-negative entries. Suppose the RSK algori thm maps $A$ to $(P,Q)$ then the RSK algorithm maps $A^T$ to $(Q,P)$, where $A^T$ is the transpose of <ma th>A[/itex]. ===Symmetric Matrices<ref name=Stanley/>=== Let $A$ be an matrix with non-negative entries, then $A=A^T</mat h> if and only if [itex]P = Q$ where $A$ is mapped to $(P, Q)$ by the RSK algorithm. ==References== {{Reflist}}