structural induction

最近在看programming languges theory相关的书，遇到了structural induction这个概念，没什么新东西，只是作为笔记记录一下。在介绍structural induction之前，先介绍一个在自然数上应用数学归纳法的例子。

定义一个集合 $S$。

• 1 $\in S$
• $3\times t+2\in S$，其中 $t\in S$
• 其它自然数都不属于 $S$

证明对于任意 $t\in S$，$t$是奇数。整个证明过程如下，

1. 首先证明base case是奇数，1显然是奇数
2. 证明“奇数”这个属性能够在推导出新值的过程中保持，$t$奇数，$3\times t$是奇数，所以$3\times t+2$显然也、 是奇数。

所以对于对于任意 $t\in S$，$t$是奇数。其实上述这个归纳过程可以说是structural induction的特例，其中Help explaining Structural Induction有一个更好的例子来说明这个问题。下面分别介绍三本书中关于structural induction的例子。

Types and Programming Languages

这本书中structural induction是在介绍terms的概念时引入的，定义

DEFINITION: The set of constans appearing in a term t, written Consts(t), is defined as follows:
\qquad
$Consts$(true) = {true}
$Consts$(false) = {false}
$Consts$(0) = {0}
$Consts$(succ t₁) = Consts(t₁)
$Consts$(pred t₁) = Consts(t₁)
$Consts$(iszero t₁) = Consts(t₁)
$Consts$(if t₁ then t₂ else t₃) = Consts(t₁) $\cup$ Consts(t₂) $\cup$ Consts(t₃)

DEFINITION: The size of a term t, writtern size(t), is defined as follows:
$size$(true) = 1
$size$(false) = 1
$size$(0) = 1
$size$(succ t₁) = $size$(t₁) + 1
$size$(pred t₁) = $size$(t₁) + 1
$size$(iszero t₁) = $size$(t₁) + 1
$size$(if t₁ then t₂ else t₃) = $size$(t₁) + $size$(t₂) + $size$(t₃) + 1

请证明The number of distinct constans in a term t is no greater then the size of t (i.e. |$Consts$(t)| $\le$ |$size$(t)|)，其实这个证明和自然数证明很相似，都是分为两步，证明基础情形$P$(t)成立，然后证明在构建term的不同case中$P$(t)能够保持。

这本书给出了在$term$上进行induction的通用形式，
$Proof$: By induction on t.

• Case: t = true
  …Show $P$(true)
• Case: t = false
  …Show $P$(false)
• Cast: t = if t₁ then t₂ else t₃
  …Show $P$(if t₁ then t₂ else t₃), using $P$(t₁), $P$(t₂), $P$(t₃)
  (And similarly for the other syntatic forms.)

我们可以看到自然数的例子是structural induciton的特殊形式，构建的case只有一种$3\times t+2\in S$。但是要应用structural induciton的一个限制是整个$Set$只能由上述的构建规则构建出来。

Practical Foundations For Programming Languages

在这本书中是在介绍Abstract Syntax Trees的时候引入structural induciton概念的，因为AST是天然的树形结构，很适合使用structural induciton。这本书定义了一个只涉及到数学运算的语言，

The trees structure of ast’s provides a very useful principle of reasoning, called structural induction.
\qquad
Suppose that we wish to prove that some property $P(a)$ holds for ast’s $a$ of a given sort. To show this, it is enough to consider all the ways in which $a$ can be generated and show that the property holds in each case under the assumption that it holds for its consituent ast’s (if any).

As an example, consider a language of arithmetic expression built from numbers, addition, and multiplication. The abstract syntax of such a language consists of a single sort $Exp$ generated by these operators:

1. An operator $num[n]$ of sort $Exp$ for each $n\in \mathbb{N}$
2. Two operators, $plus$ and $times$, of sort $Exp$, each with two arguments of sort $Exp$.

例如对于表达式(2 + (3 $\times x$)，可以表示成 $plus(num[2];times(num[3];x))$。要证明关于这棵树上的断言$P(ast)$成立，可以由下面几个步骤组成：

1. The propery holds for any variable $x$ of sort $Exp$: Prove that $P(x)$
2. The property holds for any number, $num[n]$: for every $n\in \mathbb{N}$, prove that $P(num[n])$
3. Assuming that the propery holds for $a$₁ and $a$₂, prove that it holds for $plus(a_1;a_2)$ and $times(a_1;a_2)$: if $P(a_1)andP(a_2)$, then prove $P(plus(a_1;a_2))$ and $P(times(a_1;a_2))$

Because these cases exhaust all possibilities for the formation of $a$, we are assumed that $P(a)$ holds for any ast $a$ of sort $Exp$，可以看到这里整个归纳过程遵循了和上面相同的套路。

Programming Languages and Lambda Calculi

这本书中关于structural induciton的介绍更加抽象，

structural induction applies when proving a claim about a recursively-defined set. The set is well founded, of course, only if these are atomic elements in the definition; those elements consitute the base cases for induction proofs. The self-referential elements of the definition consitute the induction cases.

证明过程大同小异，和前面的没什么区别。

Programming in Standard ML

《Practical Foundations For Programming Languages》的作者Robert Harper在CMU开设的一门课程《Programming in Standard ML》涉及到了structural induciton。

The importance of induction and recursion are not limited to functions defined over the integers. Rather, the familiar concept of mathematical induction over the natural numbers is an instance of the more general notion of structural induction over values of an inductively-defined type, see https://www.cs.cmu.edu/~rwh/introsml/techniques/structur.htm.

对于ML中$list$，我们可以将其定义成如下形式：
Generalizing a bit, we may think of the type ${}^{\prime }alist$ as inductively defined by the following clauses:

1. $nil$ is a value of type ${}^{\prime }alist$
2. If $h$ is a value of type ${}^{\prime }a$, and $t$ is a value of type ${}^{\prime }a$ list, then $h::t$ is a value of type ${}^{\prime }alist$.
3. Nothing else is a value of type ${}^{\prime }alist$.
   (The third clause is sometimes called the $extremalclause$)

按照上面的定义，我们证明关于$reverse$（也就是SML中的$rev$方法）方法的一些性质。例如我们可以证明$reversel$最终会evaluate为$l$的反转，$reverse[1,2,3,4]$的结果为$[4,3,2,1]$。

fun reverse nil = nil
  | reverse (h::t) = reverse t @ [h]

整个证明过程分为两步，

第一步$reversenil$得到$nil$，成立；

第二步，证明$reverse(h::t)$会得到$h::t$的反转，由于$reverse(h::t)=reverset@[h]$，也就是$h::t$的反转。

可以看到structural induciton是可以完美应用到涉及 inductively-defined type 的属性证明中，例如程序语言。

Help explaining Structural Induction
Untyped Arithmetic Expressions
Structural induction
Inductive definitions and structural induction - an overview
structural induciton