I'm confronted with the following problem:
Let $G=(V,E)$ be a directed graph with edge costs $c:E\rightarrow \mathbb{R}$ (Negative cycles do not matter). Let $V=\{v_1,\dots,v_n\}$. For Matrices $A$ and $B$ $\in \mathbb{R}^{n \times n}$, we define a matrix product $\odot$ as follows: $A \odot B = C$ with $$c_{i,j} = \min\left\{a_{i,l}+b_{l,j}|1\leq l \leq n\right\}.$$ We write $A=A^{\odot 1}, A \odot A = A^{\odot 2}$, etc. Let $M \in \mathbb{R}^{n \times n}$ be given by $$m_{i,j} = c(v_i,v_j)$$ with $$c(v_i,v_j)=\infty \text{ if }(v_i,v_j) \not\in E$$ Interpret the values of the matrix $M^{\odot k}$ for $K\in \mathbb{N}$, $k\geq 1$.
Does anyone know this function? For what purposes it can be used?
I wrote a script to study the behaviour of this matrix $M^{\odot k}$ and tested some instances. It seems that:
If there is a negative cycle the entries $m_{i,j}$ tend to $-\infty$ $ \forall i,j$for large $k$.
If there is a no negative cycle the entries $m_{i,j}=\infty$ $ \forall i,j$ for large $k$
If there are negative edges but no negative cycles some $m_{i,j}=\infty$ and some $-\infty<m_{i,j}<\infty$ for large $k$.
Note that $\otimes$ is the usual matrix product, but $+$ is replaced with $\min$ and $\cdot $ is replaced with $+$.
The $(i,j)$ entry of $M^{\odot k}$ gives you the cost of the cheapest length $k$ path from $v_i$ to $v_j$. You can show this by induction.
In the light of this description your observations need some corrections: