"A stochastic process X0, X1, X2, ... has the Markov property if and only if for each n the conditional distribution of Xn+1 given X0...Xn is a function only of Xn."
The only difference between the above Prop. 1.1 and the statement in TFA is that TFA omits to specify it means a conditional distribution of Xt+1 given X1...Xt (as opposed to an unconditional distribution).
But, in literally the following line, TFA goes on to clarify what it means by stating the equation that the words correspond to, and the equation has the correct conditional distribution.
I don't see reason for alarm here.
But, the Markov assumption (property) says that GIVEN (excuse my emphasis -- it is just crucial here) the present, the past does not contribute more information for predicting the future. That is the past and the future may not be independent at all, but the past and the future are conditionally independent given the present.
To discuss the Markov property, desperately need the concept of conditional independence -- there is no substitute, no easier terminology or description.
Alas, conditional independence is not so easy to discuss, even in the simplest cases. And in the grown up cases, need the Radon-Nikodym theorem of measure theory and the associated machinery of conditional expectation.
Tellingly, the part I quoted never mentioned conditional independence or even conditioning.
Sorry, again, the quote was wrong. I tried to correct the quote and, thus, keep readers trying to learn from being misled.
My sympathies are with the readers trying to learn. Early in my career, I read lots of such elementary, easy to understand introductions to lots of topics in applied math. The biggie problem was, as here, the content was from not very good down to actually wrong. Later I got a first class, rock solid foundation in grad school. In particular my background in Markov processes was from a star student of E. Cinlar, a world class expert long at Princeton. And my Ph.D. dissertation was in stochastic optimal control where the Markov assumption is just crucial -- indeed, stochastic optimal control is also commonly called Markov decision processes. In addition I've applied Markov processes in some serious work in US national security.
What I'm saying here is on the center of the target. What I quoted no one should try to learn from -- it's wrong.
No. In a Markov process, the future does depend on the past, even all of the past. But what is special is that the past and the future are conditionally independent given the present. If we are not given the present, then all of the past can be relevant in predicting the future.