Control flow is oriented around the idea of imperative statements. What is the computer doing at this given time? What will it do next? What actions should it perform?
Declarative programming abstracts this away (or at least tries to). You're program is now not a set of instructions for the computer to follow, but a sort of specification for the problem. In order to be powerful enough to be Turing complete, these specifications are usually recursive.
The main difference is that a declarative program makes no promises about how it will be run. You state the logic, and the compiler/interpreter makes a program that fulfills that logic.
One of the most important aspects of a declarative description is that it's (relatively) standalone. A procedure in an imperative language says "do this thing to the current state". A declarative program says often will not depend on its context, or will make its context more explicit.
In practice, you can usually infer the control flow of a declarative program pretty easily, especially in strict languages like ML, F# or Elm. There's a pretty close correspondence between recursive declarations and loops, and between declarative functions and imperative procedures.
Things get fuzzier in a language like Haskell, where evaluation is lazy (you don't decide what gets run when). Or in Prolog, where there's an implicit backtracking search that the runtime does for you. You're not coding these things manually, you're simply specifying what the result should look like, not how you will get that result.
