Python may not stuff its academic credentials in your face, but I think it has one of the better balances between what you can use it to teach, and what you must teach to use it. (Referencing the common and IMHO justified complaint that the simplest possible Java program requires rather a lot of stuff to be taught before you fully comprehend it.)
Teach Scheme later... or perhaps, fully distinguish between academic and practical by starting with Python to get you going then transitioning to an ML or Haskell later. (That approach has a lot of appeal to me personally, and has the benefit of the student coming out with at least one useful-in-the-industry language (Python) without compromising academic utility very much.)
Lua has much more Scheme influence than Python does (while being otherwise pretty similar). Its design assumes proficiency with C, however, so, while very practical, it's probably not as good a language for people new to programming.
Now it means having immutable values, tail recursion, easy lambdas, powerful type systems, and so on. There's a clear historical progression from past to present, but there's still been changes in what the term means over time which means that different people mean different things by "functional programming" if they don't clarify.
I had the privilege of taking an SICP-based course at the U(C). The brain-stretching is critical, isn't it? Yes, there's a lot of head-slamming that goes into modern development, but that's not really the same thing.