Before getting into my answer, let me explain what I understand by a bitableau of size $n$ (I hope it is the same as what you mean). This is a pair of tableaux $(P, P')$, where each of the integers $1,\dotsc,n$ occurs exactly once, and moreover, the numbers increase along rows and columns in both $P$ and $P'$. The shape of the bitableaux is $(\mu, \mu')$ where $\mu$ and $\mu'$ are the shapes of $P$ and $P'$. Bitableaux of shape $(\mu, \mu')$ correspond to maximal chains in the Bratelli diagram of the sequence of groups $W(B_n)$ which begin at $(\emptyset, \emptyset)$ and end at $(\mu, \mu')$.
The Robinson-Schensted correspondence has a generalization - the RSK correspondence. This correspondence is between (nonnegetive) integer matrices and pairs of semistandard Young tableaux of the same shape (see Knuth's 1970 paper in PJM, or my book). If the rows of the integer matrix sum to $\lambda = (\lambda_1,\dots, \lambda_l)$, and the columns sum to $\mu = (\mu_1,\dotsc, \mu_m)$, then the correspondence associates to $A$ a pair $(P, Q)$ of semistandard young tableaux, $P$ having content $\lambda$, and $Q$ having content $\mu$.
The matrix $A$ is said to be a partial permutation if all the row and column sums are at most $1$. Now, an element $w\in W(B_n)$ may be regarded as a pair of partial permutations, $(A, B)$ such that $A + B$ is a permutation matrix as follows: $A_{ij} = 1$ if $w:j\mapsto i$, and is $0$ otherwise. $B_{ij} = 1$ if $w:j\mapsto -i$, and is $0$ otherwise.
Now apply the RSK correspondence to $A$ and to $B$, getting $(P_A, Q_A)$ and $(P_B, Q_B)$. Then $(P_A, P_B)$ and $(Q_A, Q_B)$ are both bitableaux of the same shape.
Consider for example, the element (in your notation)
$$w = \begin{pmatrix} 1 & 2 & 3 & -1 & -2 & -3\\
2 & -1 & -3 & -2 & 1 & 3 \end{pmatrix}$$
corresponds to the matrices:
$$
A = \begin{pmatrix}
0 & 0 & 0\\
1 & 0 & 0\\
0 & 0 & 0\end{pmatrix}, \quad
B = \begin{pmatrix}
0 & 1 & 0\\
0 & 0 & 0\\
0 & 0 & 1\end{pmatrix}
$$
Under the RSK correspondence $A$ maps to $P=(1)$, $Q = (2)$ (of shape $(1)$), while $B$ corresponds to $(2, 3)$ and $(1, 3)$ (of shape $(2)$). So the bitableaux pair for $w$ is
$(1), (2, 3)$ and $(2), (1, 3)$.
The relation to domino tableaux is rather interesting. Given a bipartition $(\alpha, \beta)$, let $\lambda$ denote the unique partition whose $2$-core is trivial and whose $2$-quotient is $(\alpha, \beta)$.
Then adding a box to either $\alpha$ or $\beta$ corresponds to adding a domino to $\lambda$ (see the book by James and Kerber, or the paper of Frame, Robinson and Thrall for details). The size of $\lambda$ will be twice the sum of the sizes of $\alpha$ and $\beta$. Thus we may regard a bitableau of size $n$ as a domino-tableau of size $2n$. Thus the correspondence that I described above may also be regarded as a bijection from $W(B_n)$ onto the set of pairs of domino tableaux of the same shape.
In the running example, the associated domino tableau are $P = \begin{array}{cccc}1&2&3&3\\1&2&&\end{array}$ and $Q = \begin{array}{cccc}1&1&3&3\\2&2&&\end{array}$.
Here is some Sage code for the computation:
from sage.combinat.tableau import from_chain
def matrices(w):
n = len(w)
A = Matrix(ZZ, n, n, lambda i,j: int(w[j] == i+1))
B = Matrix(ZZ, n, n, lambda i,j: int(w[j] == -i-1))
return A, B
def domino_rsk(w, dominoes=True):
n = len(w)
A, B = matrices(w)
P1, Q1 = tuple(RSK(A))
P2, Q2 = tuple(RSK(B))
if dominoes:
ch_P1 = P1.to_chain(max_entry=n)
ch_P2 = P2.to_chain(max_entry=n)
ch_Q1 = Q1.to_chain(max_entry=n)
ch_Q2 = Q2.to_chain(max_entry=n)
P = from_chain([Partition(core=[], quotient=[ch_P1[i], ch_P2[i]]) for i in range(n+1)])
Q = from_chain([Partition(core=[], quotient=[ch_Q1[i], ch_Q2[i]]) for i in range(n+1)])
return P, Q
else:
return [P1,P2], [Q1, Q2]
For the elements of $W(B_2)$, I got (in terms of bitableaux):
[1, 2] ([[[1, 2]], []], [[[1, 2]], []])
[1, -2] ([[[1]], [[2]]], [[[1]], [[2]]])
[-1, 2] ([[[2]], [[1]]], [[[2]], [[1]]])
[-1, -2] ([[], [[1, 2]]], [[], [[1, 2]]])
[2, 1] ([[[1], [2]], []], [[[1], [2]], []])
[2, -1] ([[[1]], [[2]]], [[[2]], [[1]]])
[-2, 1] ([[[2]], [[1]]], [[[1]], [[2]]])
[-2, -1] ([[], [[1], [2]]], [[], [[1], [2]]])
and in terms of domino tableaux:
[1, 2] ([[1, 2, 2], [1]], [[1, 2, 2], [1]])
[1, -2] ([[1, 2], [1, 2]], [[1, 2], [1, 2]])
[-1, 2] ([[1, 1], [2, 2]], [[1, 1], [2, 2]])
[-1, -2] ([[1, 1, 2, 2]], [[1, 1, 2, 2]])
[2, 1] ([[1], [1], [2], [2]], [[1], [1], [2], [2]])
[2, -1] ([[1, 2], [1, 2]], [[1, 1], [2, 2]])
[-2, 1] ([[1, 1], [2, 2]], [[1, 2], [1, 2]])
[-2, -1] ([[1, 1], [2], [2]], [[1, 1], [2], [2]])
This is not the same as the answer obtained by Brian Hopkins, whose dominoes have an extra $0$, and I must admit that I have not looked carefully the references he gave as yet (and am not convinced that this question should need one to read four papers in Compositio). However, if I tweak my code so that $\lambda$ has core $(1)$ and quotient $(\alpha,\beta)$, then I do seem to get his answer.
