Documentation

This is machine translation

Translated by Microsoft
Mouseover text to see original. Click the button below to return to the English verison of the page.

Note: This page has been translated by MathWorks. Please click here
To view all translated materals including this page, select Japan from the country navigator on the bottom of this page.

lazy

Adjust Markov chain state inertia

Syntax

lc = lazy(mc)
lc = lazy(mc,w)

Description

example

lc = lazy(mc) transforms the discrete-time Markov chain mc into the lazy chain lc with an adjusted state inertia.

example

lc = lazy(mc,w) applies the inertial weights w for the transformation.

Examples

collapse all

Consider this three-state transition matrix.

Create the irreducible and periodic Markov chain that is characterized by the transition matrix P.

P = [0 1 0; 0 0 1; 1 0 0];
mc = dtmc(P);

At time t = 1,..., T, mc is forced to move to another state deterministically.

Determine the stationary distribution of the Markov chain and whether its ergodic.

xFix = asymptotics(mc)
isergodic(mc)
xFix =

    0.3333    0.3333    0.3333


ans =

  logical

   0

mc is irreducible and not ergodic. As a result, mc has a stationary distribution, but it is not a limiting distribution for all initial distributions.

Show why xFix is not a limiting distribution for all initial distributions.

x0 = [1 0 0];
x1 = x0*P
x2 = x1*P
x3 = x2*P
sum(x3 == x0) == mc.NumStates
x1 =

     0     1     0


x2 =

     0     0     1


x3 =

     1     0     0


ans =

  logical

   1

The initial distribution is reached again after several steps, which implies that the subsequent state distributions cycle through the same sets of distributions indefinitely. Therefore, mc does not have a limiting distribution.

Create an lazy version of the Markov chain mc.

lc = lazy(mc)
lc.P
lc = 

  dtmc with properties:

             P: [3x3 double]
    StateNames: ["1"    "2"    "3"]
     NumStates: 3


ans =

    0.5000    0.5000         0
         0    0.5000    0.5000
    0.5000         0    0.5000

lc is a dtmc object. At time t = 1,..., T, lc flips a fair coin. It remains in its current state if the coin shows heads and transitions to another state if the coin shows tails.

Determine the stationary distribution of the lazy chain and whether its ergodic.

lcxFix = asymptotics(lc)
isergodic(lc)
lcxFix =

    0.3333    0.3333    0.3333


ans =

  logical

   1

lc and mc have the same stationary distributions, but only lc is ergodic. Therefore, the limiting distribution of lc exists and is equal to its stationary distribution.

Consider this theoretical, right-stochastic transition matrix of a stochastic process.

Create the Markov chain that is characterized by the transition matrix P.

P = [ 0   0  1/2 1/4 1/4  0   0 ;
      0   0  1/3  0  2/3  0   0 ;
      0   0   0   0   0  1/3 2/3;
      0   0   0   0   0  1/2 1/2;
      0   0   0   0   0  3/4 1/4;
     1/2 1/2  0   0   0   0   0 ;
     1/4 3/4  0   0   0   0   0 ];
mc = dtmc(P);

Plot the eigenvalues of the transition matrix on the complex plane.

figure;
eigplot(mc);
title('Original Markov Chain')

There are three eigenvalues with modulus one, which indicates that the period of mc is three.

For illustration, create lazy versions of the Markov chain mc using various inertial weights. Plot the eigenvalues of the lazy chains on separate complex planes.

w2 = 0.1;                                 % More active Markov chain
w3 = 0.9;                                 % Lazier Markov chain
w4 = [0.9 0.1 0.25 0.5 0.25 0.001 0.999]; % Laziness differs between states

lc1 = lazy(mc);
lc2 = lazy(mc,w2);
lc3 = lazy(mc,w3);
lc4 = lazy(mc,w4);

figure;
eigplot(lc1);
title('Default Laziness');

figure;
eigplot(lc2);
title('More Active Chain');

figure;
eigplot(lc3);
title('Lazier Chain');

figure;
eigplot(lc4);
title('Differing Laziness Levels');

All lazy chains have only one eigenvalue with modulus one. Therefore, they are aperiodic. The spectral gap (distance between inner and outer circle) determines the mixing time. Observe that all lazy chains take longer to mix than the original Markov chain. In particular, chains with different inertial weights than the default take longer to mix than the default lazy chain.

Input Arguments

collapse all

Discrete-time Markov chain with NumStates states and transition matrix P, specified as a dtmc object.

Inertial weights, specified as a numeric scalar or vector of length NumStates. Values must be between 0 and 1.

  • If w is a scalar, lazy applies it to all states. That is, the transition matrix of the lazy chain (lc.P) is the result of this linear transformation

    Plazy=(1w)P+wI.

    P is mc.P and I is the NumStates-byNumStates identity matrix.

  • If w is a vector, lazy applies the weights state by state (row by row).

Data Types: double

Output Arguments

collapse all

Discrete-time Markov chain, returned as a dtmc object. lc is the lazy version of mc.

More About

collapse all

Lazy Chain

A lazy version of a Markov chain has, for each state, a probability of staying in the same state equal to at least 0.5.

In a directed graph of a Markov chain, the default lazy transformation ensures self-loops on all states, eliminating periodicity. If the Markov chain is irreducible, then its lazy version is ergodic. See graphplot.

References

[1] Gallager, R.G. Stochastic Processes: Theory for Applications. Cambridge, UK: Cambridge University Press, 2013.

Introduced in R2017b

Was this topic helpful?