The fixed point combinator doesn\'t always produce the right answer given the definition:
fix f = f (fix f)
The following code does not ter
I'm not entirely qualified to talk about what the "fixpoint combinator" is, or what the "least fixed point" is, but it is possible to use a fix-esque technique to approximate certain functions.
Translating Scala by Example section 4.4 to Haskell:
sqrt' :: Double -> Double
sqrt' x = sqrtIter 1.0
where sqrtIter guess | isGoodEnough guess = guess
| otherwise = sqrtIter (improve guess)
improve guess = (guess + x / guess) / 2
isGoodEnough guess = abs (guess * guess - x) < 0.001
This function works by repeatedly "improving" a guess until we determine that it is "good enough". This pattern can be abstracted:
myFix :: (a -> a) -- "improve" the guess
-> (a -> Bool) -- determine if a guess is "good enough"
-> a -- starting guess
-> a
fixApprox improve isGoodEnough startGuess = iter startGuess
where iter guess | isGoodEnough guess = guess
| otherwise = iter (improve guess)
sqrt'' :: Double -> Double
sqrt'' x = myFix improve isGoodEnough 1.0
where improve guess = (guess + x / guess) / 2
isGoodEnough guess = abs (guess * guess - x) < 0.001
See also Scala by Example section 5.3. fixApprox can be used to approximate the fixed point of the improve function passed into it. It repeatedly invokes improve on the input until the output isGoodEnough.
In fact, you can use myFix not only for approximations, but for exact answers as well.
primeAfter :: Int -> Int
primeAfter n = myFix improve isPrime (succ n)
where improve = succ
isPrime x = null [z | z <- [2..pred x], x `rem` z == 0]
This is a pretty dumb way to generate primes, but it illustrates the point. Hm...now I wonder...does something like myFix already exist? Stop...Hoogle time!
Hoogling (a -> a) -> (a -> Bool) -> a -> a, the very first hit is until.
until p fyields the result of applyingfuntilpholds.
Well there you have it. As it turns out, myFix = flip until.