How many possibilities to arrange a rope of length $N$ between two points Consider a $n$-dimensional lattice with $M \times M \times ... \times M$ ($n$ times) discrete grid cells ($n,M$ are natural numbers). Given two arbitrary cells at position $\vec{i}=(i_1,i_2,...,i_n)$ and $\vec{j}=(j_1,j_2,...,j_n)$ that a rope with discrete length $N$ interlinks (length is defined as the number of grid cells which can be filled in this lattice). A rope of length $0$ can only reach the point where the rope started and a rope of length $1$ can only reach the neighboring cells. The rope must 
a)be connected
b)have no self-crossing, i.e. the rope cannot fill a cell twice or more
c)cannot go out of the bounds of the lattice
d)the rope can be subdivided in sections parallel to the coordinate axes; no diagonal arrangements of sections possible
How many possibilities of arrangements for a rope with length $N$ can be counted under above requirements when the vector from one endpoints of the rope to the other is given by $\vec{r}=\vec{j}-\vec{i}$?
I would be happy for any hints. 
 A: I believe the counting would be fairly tractable if we did not have to exclude self-intersections, but did have to respect the coordinate bounds $1$ to $M$ of the (product) lattice.  
This is based on comparing the situation here with the more familiar problem in which monotone progress must be made going from $\vec{i}$ to $\vec{j}$, equiv. when the path length is precisely the taxicab distance $N = \sum_k |j_k - i_k|$.  In that case there is no possibility of self-intersection, and a standard combinatorial argument counts the paths as:
$$ \text{# of paths} = \frac{N!}{|j_1-i_1|! |j_2-i_2|! \cdots |j_n-i_n|!} $$
In the present context, where the length of rope exceeds the taxicab distance (it must do so by an even number to make a path feasible), it would be fairly easy to incorporate the bounds $1$ to $M$ on each coordinate.  However the problem of self-intersections appears difficult to account for in an exact way.

Here is an equivalent representation of the feasible rope paths.  No claim is made that this leads to an efficient method of counting, but it does give a starting point to investigate symmetries of the solution space.
Consider a matrix $A$ of size $n\times N$ with entries in $\{-1,0,1\}$.  Each column contains exactly one nonzero entry, and the sum of entries in the $k$th row equals $j_k - i_k$.  Such matrices represent distinct walks from $\vec{i}$ to $\vec{j}$, subject to the constraint of length $N$.
Two further sets of restrictions are needed to satisfy the conditions imposed by the Question.  The first is that the walk should be a path, i.e. no self-intersection (equiv. no revisiting the same point twice).  This can be imposed by referring to a matrix $A'$ derived from $A$ by forming a running sum of the columns:
$$ A'(k,m) = \sum_{\ell=1}^m A(k,\ell) $$
for $k = 1,\ldots,n$.  The no self-intersection condition is then equivalent to the columns of $A'$ being distinct and nonzero.  The last is the avoidance of a return to the starting point.  Note that the last column of $A'$ must by previous construction have entries $A(k,N) = j_k - i_k$.
Finally we need to restrict the path from leaving the bounds of the lattice.  Let $\vec{i}$ be transposed into a column $\vec{i}^T$.  Then adding $\vec{i}^T$ to each column of $A'$ gives a matrix $A''$ whose columns represent the successive points visited by the rope path.  Thus staying within the bounds means for all $1\le k \le n$ and $1\le m \le N$:
$$ 1 \le A''(k,m) \le M $$
together, naturally, with requiring the starting point $\vec{i}$ to lie within the lattice bounds.
