algorithm - How does inorder+preorder construct unique binary tree?

Question

Recently, my questions were marked duplicate, like this , even if they weren't. So, let me start with following and then I'll explain my question.

Why this question is not a duplicate?

I'm not asking how to create a binary tree when inorder & preorder traversal are given. I'm asking for the proof, that inorder+preorder traversal define a unique binary tree.

Now, to original question. I went to an interview and interviewer asked me this question. I was stuck and couldn't proceed. :|

Question: Given inorder & preorder traversal of a binary tree. Prove there is only one binary tree possible with given data. In other words, prove two different binary trees can't have same inorder & preorder traversal. Assume all the elements in the tree are unique (thanks to @envy_intelligence for pointing this assumption).

I tried convincing interviewer using examples, but interviewer was asking for mathematical/intuitive proof. Can anyone help me proving it?

Answer 1

Start with the preorder traversal. Either it is empty, in which case you are done, or it has a first element, r0, the root of the tree. Now search the inorder traversal for r0. The left subtree will all come before that point and the right subtree will all come after that point. Thus you can divide the inorder traversal at that point into an inorder traversal of the left subtree il and an inorder traversal of the right subtree, ir.

If il is empty, then the rest of the preorder traversal belongs to the right subtree, and you can continue inductively. If ir is empty, the same thing happens on the other side. If neither is empty, find the first element of ir in the remainder of the preorder traversal. This divides it into a preorder traversal of the left subtree and one of the right subtree. Induction is immediate.

In case anyone is interested in a formal proof, I have (finally) managed to produce one in Idris. I have not, however, taken the time to try to make it terribly readable, so it is actually fairly hard to read much of it. I would recommend that you look mostly at the top-level types (i.e., lemmas, theorems, and definitions) and try to avoid getting too bogged down in the proofs (terms).

First some preliminaries:

module PreIn
import Data.List
%default total

Now the first real idea: a binary tree.

data Tree : Type -> Type where
  Tip : Tree a
  Node : (l : Tree a) -> (v : a) -> (r : Tree a) -> Tree a
%name Tree t, u

Now the second big idea: the idea of a way to find a particular element in a particular tree. This is based closely on the Elem type in Data.List, which expresses a way to find a particular element in a particular list.

data InTree : a -> Tree a -> Type where
  AtRoot : x `InTree` (Node l x r)
  OnLeft : x `InTree` l -> x `InTree` (Node l v r)
  OnRight : x `InTree` r -> x `InTree` (Node l v r)

Then there are a whole slew of horrible lemmas, a couple of which were suggested by Eric Mertens (glguy) in his answer to my question about it.

Horrible lemmas

size : Tree a -> Nat
size Tip = Z
size (Node l v r) = size l + (S Z + size r)

onLeftInjective : OnLeft p = OnLeft q -> p = q
onLeftInjective Refl = Refl

onRightInjective : OnRight p = OnRight q -> p = q
onRightInjective Refl = Refl

inorder : Tree a -> List a
inorder Tip = []
inorder (Node l v r) = inorder l ++ [v] ++ inorder r

instance Uninhabited (Here = There y) where
  uninhabited Refl impossible

instance Uninhabited (x `InTree` Tip) where
  uninhabited AtRoot impossible

elemAppend : {x : a} -> (ys,xs : List a) -> x `Elem` xs -> x `Elem` (ys ++ xs)
elemAppend [] xs xInxs = xInxs
elemAppend (y :: ys) xs xInxs = There (elemAppend ys xs xInxs)

appendElem : {x : a} -> (xs,ys : List a) -> x `Elem` xs -> x `Elem` (xs ++ ys)
appendElem (x :: zs) ys Here = Here
appendElem (y :: zs) ys (There pr) = There (appendElem zs ys pr)

tThenInorder : {x : a} -> (t : Tree a) -> x `InTree` t -> x `Elem` inorder t
tThenInorder (Node l x r) AtRoot = elemAppend _ _ Here
tThenInorder (Node l v r) (OnLeft pr) = appendElem _ _ (tThenInorder _ pr)
tThenInorder (Node l v r) (OnRight pr) = elemAppend _ _ (There (tThenInorder _ pr))

listSplit_lem : (x,z : a) -> (xs,ys:List a) -> Either (x `Elem` xs) (x `Elem` ys)
  -> Either (x `Elem` (z :: xs)) (x `Elem` ys)
listSplit_lem x z xs ys (Left prf) = Left (There prf)
listSplit_lem x z xs ys (Right prf) = Right prf


listSplit : {x : a} -> (xs,ys : List a) -> x `Elem` (xs ++ ys) -> Either (x `Elem` xs) (x `Elem` ys)
listSplit [] ys xelem = Right xelem
listSplit (z :: xs) ys Here = Left Here
listSplit {x} (z :: xs) ys (There pr) = listSplit_lem x z xs ys (listSplit xs ys pr)

mutual
  inorderThenT : {x : a} -> (t : Tree a) -> x `Elem` inorder t -> InTree x t
  inorderThenT Tip xInL = absurd xInL
  inorderThenT {x} (Node l v r) xInL = inorderThenT_lem x l v r xInL (listSplit (inorder l) (v :: inorder r) xInL)

  inorderThenT_lem : (x : a) ->
                     (l : Tree a) -> (v : a) -> (r : Tree a) ->
                     x `Elem` inorder (Node l v r) ->
                     Either (x `Elem` inorder l) (x `Elem` (v :: inorder r)) ->
                     InTree x (Node l v r)
  inorderThenT_lem x l v r xInL (Left locl) = OnLeft (inorderThenT l locl)
  inorderThenT_lem x l x r xInL (Right Here) = AtRoot
  inorderThenT_lem x l v r xInL (Right (There locr)) = OnRight (inorderThenT r locr)

unsplitRight : {x : a} -> (e : x `Elem` ys) -> listSplit xs ys (elemAppend xs ys e) = Right e
unsplitRight {xs = []} e = Refl
unsplitRight {xs = (x :: xs)} e = rewrite unsplitRight {xs} e in Refl

unsplitLeft : {x : a} -> (e : x `Elem` xs) -> listSplit xs ys (appendElem xs ys e) = Left e
unsplitLeft {xs = []} Here impossible
unsplitLeft {xs = (x :: xs)} Here = Refl
unsplitLeft {xs = (x :: xs)} {ys} (There pr) =
  rewrite unsplitLeft {xs} {ys} pr in Refl

splitLeft_lem1 : (Left (There w) = listSplit_lem x y xs ys (listSplit xs ys z)) ->
                 (Left w = listSplit xs ys z) 

splitLeft_lem1 {w} {xs} {ys} {z} prf with (listSplit xs ys z)
  splitLeft_lem1 {w}  Refl | (Left w) = Refl
  splitLeft_lem1 {w}  Refl | (Right s) impossible

splitLeft_lem2 : Left Here = listSplit_lem x x xs ys (listSplit xs ys z) -> Void
splitLeft_lem2 {x} {xs} {ys} {z} prf with (listSplit xs ys z)
  splitLeft_lem2 {x = x} {xs = xs} {ys = ys} {z = z} Refl | (Left y) impossible
  splitLeft_lem2 {x = x} {xs = xs} {ys = ys} {z = z} Refl | (Right y) impossible

splitLeft : {x : a} -> (xs,ys : List a) ->
            (loc : x `Elem` (xs ++ ys)) ->
            Left e = listSplit {x} xs ys loc ->
            appendElem {x} xs ys e = loc
splitLeft {e} [] ys loc prf = absurd e
splitLeft (x :: xs) ys Here prf = rewrite leftInjective prf in Refl
splitLeft {e = Here} (x :: xs) ys (There z) prf = absurd (splitLeft_lem2 prf)
splitLeft {e = (There w)} (y :: xs) ys (There z) prf =
  cong $ splitLeft xs ys z (splitLeft_lem1 prf)

splitMiddle_lem3 : Right Here = listSplit_lem y x xs (y :: ys) (listSplit xs (y :: ys) z) ->
                   Right Here = listSplit xs (y :: ys) z

splitMiddle_lem3 {y} {x} {xs} {ys} {z} prf with (listSplit xs (y :: ys) z)
  splitMiddle_lem3 {y = y} {x = x} {xs = xs} {ys = ys} {z = z} Refl | (Left w) impossible
  splitMiddle_lem3 {y = y} {x = x} {xs = xs} {ys = ys} {z = z} prf | (Right w) =
    cong $ rightInjective prf  -- This funny dance strips the Rights off and then puts them
                               -- back on so as to change type.


splitMiddle_lem2 : Right Here = listSplit xs (y :: ys) pl ->
                   elemAppend xs (y :: ys) Here = pl

splitMiddle_lem2 {xs} {y} {ys} {pl} prf with (listSplit xs (y :: ys) pl) proof prpr
  splitMiddle_lem2 {xs = xs} {y = y} {ys = ys} {pl = pl} Refl | (Left loc) impossible
  splitMiddle_lem2 {xs = []} {y = y} {ys = ys} {pl = pl} Refl | (Right Here) = rightInjective prpr
  splitMiddle_lem2 {xs = (x :: xs)} {y = x} {ys = ys} {pl = Here} prf | (Right Here) = (Refl impossible) prpr
  splitMiddle_lem2 {xs = (x :: xs)} {y = y} {ys = ys} {pl = (There z)} prf | (Right Here) =
    cong $ splitMiddle_lem2 {xs} {y} {ys} {pl = z} (splitMiddle_lem3 prpr)

splitMiddle_lem1 : Right Here = listSplit_lem y x xs (y :: ys) (listSplit xs (y :: ys) pl) ->
                   elemAppend xs (y :: ys) Here = pl

splitMiddle_lem1 {y} {x} {xs} {ys} {pl} prf with (listSplit xs (y :: ys) pl) proof prpr
  splitMiddle_lem1 {y = y} {x = x} {xs = xs} {ys = ys} {pl = pl} Refl | (Left z) impossible
  splitMiddle_lem1 {y = y} {x = x} {xs = xs} {ys = ys} {pl = pl} Refl | (Right Here) = splitMiddle_lem2 prpr

splitMiddle : Right Here = listSplit xs (y::ys) loc ->
              elemAppend xs (y::ys) Here = loc

splitMiddle {xs = []} prf = rightInjective prf
splitMiddle {xs = (x :: xs)} {loc = Here} Refl impossible
splitMiddle {xs = (x :: xs)} {loc = (There y)} prf = cong $ splitMiddle_lem1 prf

splitRight_lem1 : Right (There pl) = listSplit (q :: xs) (y :: ys) (There z) ->
                  Right (There pl) = listSplit xs (y :: ys) z

splitRight_lem1 {xs} {ys} {y} {z} prf with (listSplit xs (y :: ys) z)
  splitRight_lem1 {xs = xs} {ys = ys} {y = y} {z = z} Refl | (Left x) impossible
  splitRight_lem1 {xs = xs} {ys = ys} {y = y} {z = z} prf | (Right x) =
    cong $ rightInjective prf  -- Type dance: take the Right off and put it back on.

splitRight : Right (There pl) = listSplit xs (y :: ys) loc ->
             elemAppend xs (y :: ys) (There pl) = loc
splitRight {pl = pl} {xs = []} {y = y} {ys = ys} {loc = loc} prf = rightInjective prf
splitRight {pl = pl} {xs = (x :: xs)} {y = y} {ys = ys} {loc = Here} Refl impossible
splitRight {pl = pl} {xs = (x :: xs)} {y = y} {ys = ys} {loc = (There z)} prf =
  let rec = splitRight {pl} {xs} {y} {ys} {loc = z} in cong $ rec (splitRight_lem1 prf)

Correspondence between a tree and its inorder traversal

These horrible lemmas lead up to the following theorems about inorder traversals, which together demonstrate a one-to-one correspondence between ways to find a particular element in a tree and ways to find that element in its inorder traversal.

---------------------------
-- tThenInorder is a bijection from ways to find a particular element in a tree
-- and ways to find that element in its inorder traversal. `inorderToFro`
-- and `inorderFroTo` together demonstrate this by showing that `inorderThenT` is
-- its inverse.

||| `tThenInorder t` is a retraction of `inorderThenT t`
inorderFroTo : {x : a} -> (t : Tree a) -> (loc : x `Elem` inorder t) -> tThenInorder t (inorderThenT t loc) = loc
inorderFroTo Tip loc = absurd loc
inorderFroTo (Node l v r) loc with (listSplit (inorder l) (v :: inorder r) loc) proof prf
  inorderFroTo (Node l v r) loc | (Left here) =
    rewrite inorderFroTo l here in splitLeft _ _ loc prf
  inorderFroTo (Node l v r) loc | (Right Here) = splitMiddle prf
  inorderFroTo (Node l v r) loc | (Right (There x)) =
    rewrite inorderFroTo r x in splitRight prf

||| `inorderThenT t` is a retraction of `tThenInorder t`
inorderToFro : {x : a} -> (t : Tree a) -> (loc : x `InTree` t) -> inorderThenT t (tThenInorder t loc) = loc
inorderToFro (Node l v r) (OnLeft xInL) =
  rewrite unsplitLeft {ys = v :: inorder r} (tThenInorder l xInL)
  in cong $ inorderToFro _ xInL
inorderToFro (Node l x r) AtRoot =
  rewrite unsplitRight {x} {xs = inorder l} {ys = x :: inorder r} (tThenInorder (Node Tip x r) AtRoot)
  in Refl
inorderToFro {x} (Node l v r) (OnRight xInR) =
  rewrite unsplitRight {x} {xs = inorder l} {ys = v :: inorder r} (tThenInorder (Node Tip v r) (OnRight xInR))
  in cong $ inorderToFro _ xInR

Correspondence between a tree and its preorder traversal

Many of those same lemmas can then be used to prove the corresponding theorems for preorder traversals:

preorder : Tree a -> List a
preorder Tip = []
preorder (Node l v r) = v :: (preorder l ++ preorder r)

tThenPreorder : (t : Tree a) -> x `InTree` t -> x `Elem` preorder t
tThenPreorder Tip AtRoot impossible
tThenPreorder (Node l x r) AtRoot = Here
tThenPreorder (Node l v r) (OnLeft loc) = appendElem _ _ (There (tThenPreorder _ loc))
tThenPreorder (Node l v r) (OnRight loc) = elemAppend (v :: preorder l) (preorder r) (t