Bellman equation derivation: Difference between revisions

From Machinelearning
No edit summary
No edit summary
Line 12: Line 12:


So the law of total probability states that <math display="inline">\Pr(B \mid A) = \sum_{j=1}^n \Pr(B \mid C_j,A)\Pr(C_j\mid A)</math>.
So the law of total probability states that <math display="inline">\Pr(B \mid A) = \sum_{j=1}^n \Pr(B \mid C_j,A)\Pr(C_j\mid A)</math>.
Now we see how the law of total probability interacts with conditional expectation. Let <math>X</math> be a random variable. Then <math>\mathbb E[X \mid A] = \sum_x x \cdot \Pr(X = x \mid A) = \sum_x x \cdot \sum_j \Pr(X =x \mid C_j,A)\Pr(C_j\mid A)</math>. Here the event <math>X=x</math> is playing the role of <math>B</math> in the statement of the conditional law of total probability.

Revision as of 00:58, 1 September 2019

Bellman equation for .

We want to show for all states .

The core idea of the proof is to use the law of total probability to go from marginal to conditional probabilities, and then invoke the Markov assumption.

The law of total probability states that if is an event, and are events that partition the sample space, then .

For fixed event with non-zero probability, the mapping is another valid probability measure. In other words, define by for all events . Now the law of total probability for states that . We also have

So the law of total probability states that .

Now we see how the law of total probability interacts with conditional expectation. Let be a random variable. Then . Here the event is playing the role of in the statement of the conditional law of total probability.