Monotonicity and Convexity of Stochastic Matrices

686 Views Asked by At

The definition of stochastic monotonicity and convexity is given by "Stochastic Orders and Their Applications" by Moshe Shaked and George Shanthikumar (1994) as:

Let $P = \{p_{i,j} \}$ be a stochastic matrix on $\mathbb{N}_+$. Then P is said to be

1) Stochastically monotone if $\sum_{j=k}^\infty p_{i,j} \leq \sum_{j=k}^\infty p_{(i+1),j}$ for any $i$ and $k$ in $\mathbb{N}_+$;

2) Stochastically monotone convex $\sum_{j=k}^\infty p_{i,j} \leq \sum_{j=k+1}^\infty p_{(i+1),j}$ for any $i$ and $k$ in $\mathbb{N}_+$;

That is, the row vectors of the matrix are stochastically increasing. Many results follow from this definition. For example, one can show stochastic ordering between two transition matrices of discrete-time Markov chains. This in turn, implies that their respective steady-state probability distributions are ordered. I want to show that similar properties hold for the steady-state probabilities in my system. However, my problem is defined slightly differently.

Let $Q = \{q_{i,j} \}$ be my stochastic matrix on $\mathbb{N}_+$. In my system, the following relations hold:

1) $\sum_{j=k}^\infty q_{i,j} \geq \sum_{j=k}^\infty q_{(i+1),j}$ for any $i$ and $k$ in $\mathbb{N}_+$;

2) $\sum_{j=k}^\infty q_{i,j} \geq \sum_{j=k+1}^\infty q_{(i+1),j}$ for any $i$ and $k$ in $\mathbb{N}_+$;

Note that the $\leq$ sign in the original definition is replaced by a $\geq$ sign in my system. I have tried redefining the underlying state space, using a coupling argument, and tried to find a bounding matrix that satisfies stochastic monotonicity and convexity but also has the same steady-state distribution as the one induced by Q, all to no avail. If anybody has any ideas and/or could provide me with some help in this matter it would be much appreciated.

1

There are 1 best solutions below

5
On

If some $Q$ satisfies your property (1) then $Q^2$ is stochastically monotone.