Saturday, May 10, 2014

Well-ordering and induction

$\Newextarrow{\xleftrightarrow}{10,10}{0x2194} \Newextarrow{\xRightarrow}{10,10}{0x21D2} \Newextarrow{\xLeftrightarrow}{10,10}{0x21D4}$ 2018-06-20 to ???

ABSTRACT: We define the four vertices and prove the seven implications of: \[\begin{array}{} && WO && \longrightarrow \atop \longleftarrow && DCC \\ & \swarrow && \searrow\nwarrow \\ SIND && \longrightarrow \atop \longleftarrow && PIND \mkern30mu . \\ \end{array}\] The proofs involve an interesting combination of first- and second-order logic.



Context:

$\N = \{0,1,2,\ldots\}$ is the set of natural numbers, and $\functions : \N \longrightarrow \N : \numbern \mapsto \numbern+1$ is its successor endofunction.
Let $\N$ also denote the corresponding discrete category, whose only arrows are its identity arrows, and let $\functions$ also denote the corresponding endofunctor.
Let $\bftwo$ be the category whose objects are $\leftadj\bot$ and $\rightadj\top$, whose only non-identity arrow is $\leftadj\bot \lt \rightadj\top$, also notated $\leftadj\bot \to \rightadj\top$ or $\leftadj\bot \Rightarrow \rightadj\top$
(an arrow realizing the ex falso sequitur quodlibet principle of logic).
Functions/functors $\functionA:\N \longrightarrow \bftwo : \numbern\mapsto\functionA_\numbern$ are called predicates on $\N$.



$ICON$, $SCON$, and $SIND$

The two conditions (on a predicate $\functionA:\N \rightarrow \bftwo$) for successor induction are:

initial condition ($ICON$):
$\functionA_0$ is true, i.e., there is an arrow in $\bftwo$, $ICON : \rightadj\top \xRightarrow{\textstyle ICON} \functionA_0$, i.e., there is a 2-cell in $\Cat$: $\mkern20mu$ \(\begin{array}{} \catI && { \xrightarrow[]{\textstyle 0} } && \N \\ & \llap{\rightadj\top} \searrow & \xRightarrow{\textstyle ICON} & \swarrow \rlap{\lower3pt\hbox{$\mkern-10mu\functionA$}} \\ && \bftwo \\ \end{array}\)
successor condition ($SCON$):
$SCON : (\forall\numbern\in\N) \Big[ \functionA_\numbern \xRightarrow{\textstyle SCON(\numbern)} \functionA_{\numbern+1} \Big]$
i.e., there is a transformation (automatically natural because $\N$ is discrete) in $\Cat$: \[\begin{array}{} &&&&&& \N \\ { \N } && { \xrightarrow[]{\textstyle \mkern10mu \functionA \mkern10mu} } && \bftwo &&& \mkern10mu \searrow \rlap{\raise3pt\hbox{$\mkern-10mu\functionA$}} &&&& \N && \xrightarrow{\textstyle \functions} && \N \\ & \llap{\lower3pt\hbox{$\functions \mkern{-10mu}$}} \searrow & \Big\Downarrow \rlap{\mkern-34mu SCON} & \nearrow \rlap{\lower3pt\hbox{$\mkern-10mu\functionA$}} && \mskip80mu \text{i.e.} \mskip80mu & \llap{\functions} \Bigg\downarrow & \mkern-20mu\Big\Downarrow\rlap{\mkern-34mu SCON} && \bftwo & \mskip80mu \text{i.e.} \mskip80mu && \llap{\functionA}\searrow & \xRightarrow{\textstyle SCON} & \swarrow\rlap{\functionA \mskip 60mu ,} \\ && \N &&&&& \mkern10mu \nearrow \rlap{\lower3pt\hbox{$\mkern-10mu\functionA$}} &&&&&& \bftwo \\ &&&&&& \N \\ \end{array}\] i.e., $\functorA$ is an opcoalgebra (equivalently, a co-opalgebra) for the endofunction/endofunctor $\functions$, i.e., an $\functions$-opcoalgebra or $\functions$-co-opalgebra.

The (initial condition $ICON$) and (successor condition $SCON$) form the antecedent in (successor induction $SIND$):

successor induction ($SIND$):
$SIND : (ICON \wedge SCON) \xRightarrow{\textstyle SIND} (\functionA = \top)$, i.e.,
$SIND : \bigg( \functionA_0 \wedge (\forall\numbern\in\N) \Big[ \functionA_\numbern \xRightarrow{\textstyle SCON(\numbern)} \functionA_{\numbern+1} \Big] \bigg) \xRightarrow{\textstyle SIND} (\functionA = \top)$.
This is generally called “the principle of mathematical induction” or “induction schema”,
however calling it “successor induction” makes more clear its relation to the “predecessors induction” (often called “complete induction”) defined below.



The predicate $\functionA$ generates a 2-diagram in $\Cat$:

The above data generate a 2-diagram in the 2-category $\Cat$, shown twice below
(where $\numbern_0$ and $\numbern_0-1$ are hypothetical elements of $\N$, and $\catI$ is the unit category):

\[\begin{array}{} \numbern_0-1 & \in & { \rightadj\top\downarrow\functionA } && { \xrightarrow[]{\textstyle \mkern40mu \llap{! \mkern20mu} {} \mkern40mu} } && \catI \\ && & \searrow & \raise6pt\hbox{$\Vert$} & { \nearrow \rlap{\lower3pt\hbox{${!}$}} } \\ && && { \rightadj\top\downarrow\functionA } \\ && { \Big\downarrow } && { \mkern{-50mu} \Big\Downarrow } && \rightadj{ \Big\downarrow \rlap\top } \\ && & \Big\Vert && { \Big\Downarrow } \\ && && { \Big\downarrow } \\ && { \N } && { \xrightarrow[]{\textstyle \mkern40mu \llap{\functionA\mkern20mu} {} \mkern40mu} } && \bftwo \\ & \llap{\raise3pt\hbox{$\numbern_0 - 1 \mkern{-10mu}$}} \nearrow & \Vert & \llap{\lower3pt\hbox{$\functions \mkern{-10mu}$}} \searrow & \Big\Downarrow \rlap{\mkern-34mu SCON} & { \nearrow \rlap{\lower3pt\hbox{$\mkern-10mu\functionA$}} } & \leftadj{ \Big\uparrow \rlap\bot } \\ \catI && \xrightarrow[\textstyle \numbern_0]{} && { \N } & \Big\Downarrow & \catI \\ &&&& \Big\uparrow & { \nearrow \rlap{\lower3pt\hbox{${!}$}} } \\ && \numbern_0 & \in & \functionA\downarrow\leftadj\bot \\ \end{array}\] The same diagram rotated clockwise by a quarter-turn:
(you have to add two long southwest-pointing diagonals yourself,
instances of $\functionA$ and of ${!}$,
and add three 2-cells, corresponding to the
lower ($SCON$),
back (universal), and
top (equality)
faces of the triangular prism in the diagram on the left).
\[\begin{array}{} && \catI \\ &&& \searrow \rlap{\numbern_0-1} \\ && \llap{\numbern_0} \Bigg\downarrow & = & \N & \leftarrow & \rightadj\top\downarrow\functionA \\ &&& \swarrow \rlap{\functions} & = & \swarrow\\ \functionA\downarrow\leftadj\bot & \xrightarrow{} & \N & \xleftarrow{} & \rightadj\top\downarrow\functionA \\ \llap{!} \Bigg\downarrow & \Leftarrow & \Bigg\downarrow \rlap{\mkern-14mu\functionA} & \Leftarrow & \Bigg\downarrow \rlap{!} \\ \catI & \xrightarrow[\textstyle \leftadj\bot]{} & \bftwo & \xleftarrow[\textstyle \rightadj\top]{} & \catI \\ \end{array}\]


Comparing (predicates on $\N$) and (subsets of $\N$)

In comparing induction, which is about (predicates on $\N$), to well-ordering, which is about (subsets of $\N$),
we will use a certain bijection between the two, the composite shown below: \[\begin{array}{} [\N,\bftwo] & \xleftrightarrow{\text{characteristic map}} & \Sub(\N) & \xleftrightarrow{\text{set-theoretic complement}} & \Sub(\N) \\ \functionA & \leftrightarrow & \top\downarrow\functionA & \leftrightarrow & \functionA\downarrow\bot \\ {?}\notin\setS = {?}\in\tilde\setS & \leftrightarrow & \tilde\setS & \leftrightarrow & \setS \\ \end{array}\]



$WO \Rightarrow SIND$

Suppose ($\functionA\downarrow\leftadj\bot$ is non-empty), i.e., ($\functionA\downarrow\leftadj\bot \not= \emptyset$), i.e., $(\exists\numbern\in\N)(\numbern\in\functionA\downarrow\leftadj\bot)$, i.e., $(\exists\numbern\in\N)(\functionA_\numbern \to \leftadj\bot)$, i.e., ($\functionA_\numbern$ is false for some $\numbern\in\N$).
If ($\N$ is well-ordered), then (the non-empty $\functionA\downarrow\leftadj\bot$) has (a least member, $\boxed{\numbern_0}$),
i.e., $(\numbern_0 \in \functionA\downarrow\leftadj\bot)$ and $(\forall \numbern \in \functionA\downarrow\leftadj\bot) (\numbern_0 \le \numbern)$.
Then we have \[\top \Rightarrow (\top \xRightarrow{ICON} \functionA_0) \Rightarrow (0 \in \rightadj\top\downarrow\functionA) \xRightarrow{\cap = \emptyset} (0 \notin \functionA\downarrow\leftadj\bot) \xRightarrow{\numbern_0 \in \functionA\downarrow\leftadj\bot} ( \numbern_0 \not= 0 ) \Rightarrow (\exists (\numbern_0 - 1) \in \N) \; . \] Then \[\top \Rightarrow (\numbern_0-1 \lt \numbern_0) \xRightarrow{\cap} (\numbern_0 \not\le \numbern_0-1) \xRightarrow[\text{contrapositive}]{\numbern_0 \text{ least}} (\numbern_0-1 \notin \functionA\downarrow\leftadj\bot) \xRightarrow{\cup} (\numbern_0-1 \in \rightadj\top\downarrow\functionA) \xRightarrow{SCON(\numbern_0 - 1)} ((\numbern_0-1)+1 = \numbern_0 \in \rightadj\top\downarrow\functionA) \xRightarrow{\cap} (\numbern_0 \notin \functionA\downarrow\leftadj\bot) \xRightarrow{\numbern_0 \in \functionA\downarrow\leftadj\bot} \bot \; . \] We have derived a contradiction, so one of our assumptions must be invalid.
If $\N$ is well-ordered and $\functionA$ satisfies $ICON$ and $SCON(\numbern_0 - 1)$,
this must mean that ($\functionA\downarrow\leftadj\bot$ is NOT non-empty).
By (the law of the excluded middle),
this means ($\functionA\downarrow\leftadj\bot$ is empty), i.e., (its complement $\rightadj\top\downarrow\functionA$ is all of $\N$).
So ($\functionA_\numbern$ is true for all $\numbern\in\N$).
Expressing this symbolically, and deriving the desired consequence, we have: \[\begin{array}{} WO \wedge (\functionA\downarrow\bot \not= \emptyset) \wedge ICON \wedge SCON \xRightarrow{WO} \big( (\exists\numbern_0\in\functionA\downarrow\bot)(\forall \numbern \in \functionA\downarrow\bot) (\numbern_0 \le \numbern) \big) \wedge ICON \wedge SCON \xRightarrow{\text{by the above}} \bot \\ WO \wedge ICON \wedge SCON \Longrightarrow \big[ \neg (\functionA\downarrow\bot \not= \emptyset) \big] \xlongequal{\text{excluded middle}} (\functionA\downarrow\bot = \emptyset) \xlongequal{\cup} (\top\downarrow\functionA = \N) = (\functionA=\top), \rlap{\mkern30mu \text{in short:}} \\ (WO \wedge ICON \wedge SCON) \Longrightarrow (\functionA = \top) \\ WO \Longrightarrow \big[ (ICON\wedge SCON) \Rightarrow (\functionA = \top) \big] = SIND \; .\\ \end{array}\]



$PCON$ and $PIND$

There is an alternative version of induction,
which replaces the two conditions, the initial condition ($ICON$) and the successor condition ($SCON$) of successor induction ($SIND$),
with a single induction condition, the predecessors condition ($PCON$) (note the plural):

predecessors condition ($PCON$):
$PCON : (\forall\numbern\in\N)\Big[(\forall\numberk\lt\numbern)\functionA_\numberk \xRightarrow[]{\textstyle PCON(\numbern)} \functionA_\numbern \Big]$, $\mkern40mu$ i.e., by the definition of relative quantification,
$PCON : (\forall\numbern\in\N)\Big[(\forall\numberk\in\N)[\numberk\lt\numbern \Rightarrow \functionA_\numberk] \xRightarrow[]{\textstyle PCON(\numbern)} \functionA_\numbern \Big]$.
In interpreting those logical statements, recall that the quantifiers “$\forall$” and “$\exists$” bind more tightly than binary connectives such as “$\Rightarrow$”,
so $(\forall\numberk\lt\numbern)\functionA_\numberk \Rightarrow \functionA_\numbern$ is equivalent to $\big((\forall\numberk\lt\numbern)\functionA_\numberk \big) \Rightarrow \functionA_\numbern$, NOT $(\forall\numberk\lt\numbern)[\functionA_\numberk \Rightarrow \functionA_\numbern]$.

(The predecessors condition for $\numbern=0$, $PCON(0)$), is equivalent to (the initial condition, $ICON$),
for the following are logically equivalent:

\(\begin{array}{r} PCON(0) \\ (\forall\numberk\lt 0)\functionA_\numberk \Rightarrow \functionA_0 \\ (\forall\numberk\in\N)[\numberk\lt 0 \Rightarrow \functionA_\numberk] \Rightarrow \functionA_0 \\ (\forall\numberk\in\N)[\bot \Rightarrow \functionA_\numberk] \Rightarrow \functionA_0 \\ (\forall\numberk\in\N)(\top) \Rightarrow \functionA_0 \\ \top \Rightarrow \functionA_0 \\ ICON \\ \end{array}\)

(The equivalence $(\forall\numberk\lt 0)\functionA_\numberk \Leftrightarrow \top$ demonstrated above) is sometimes described by (the phrase “$(\forall\numberk\lt 0)\functionA_\numberk$ is vacuously true”).

(The predecessors condition $PCON$) is the antecedent in (predecessors induction $PIND$):

predecessors induction ($PIND$):
$PIND : PCON \xRightarrow{\textstyle PIND} (\functionA = \top)$ , $\mkern40mu$ i.e.,
$PIND : (\forall\numbern\in\N)\Big[(\forall\numberk\lt\numbern)\functionA_\numberk \xRightarrow[]{\textstyle PCON(\numbern)} \functionA_\numbern \Big] \xRightarrow{\textstyle PIND} (\functionA = \top)$ .


$WO \Rightarrow PIND$

The proof above that $WO \Rightarrow SIND$ can be easily modified to show $WO \Rightarrow PIND$.
The key is to examine what it means to say that ($\numbern_0$ is least in $\functionA\downarrow\leftadj\bot$).
On the one hand, ($\numbern_0\in \functionA\downarrow\leftadj\bot$) means ($\functionA_{\numbern_0}$ is false).
On the other hand, we have a chain of logical equivalences:

\(\begin{array}{l} (\forall\numbern\in \functionA\downarrow\leftadj\bot) (\numbern_0\le\numbern)\\ (\forall\numbern\in\N) [\numbern\in \functionA\downarrow\leftadj\bot \Rightarrow \numbern_0\le\numbern] & \text{ (definition of relative quantifier)} \\ (\forall\numbern\in\N) [\numbern_0\not\le\numbern \Rightarrow \numbern\notin \functionA\downarrow\leftadj\bot] & \text{ (taking the contrapositive)} \\ (\forall\numbern\in\N) [\numbern\lt\numbern_0 \Rightarrow \numbern\in \rightadj\top\downarrow\functionA] & \text{ (the predicates $({?}\in\functionA\downarrow\leftadj\bot)$ and $(\numbern_0\le {?})$ each yield a partition of $\N$)} \\ (\forall\numberk\in\N) [\numberk\lt\numbern_0 \Rightarrow \functionA_\numberk] & \text{ (change of variable $\numbern\mapsto\numberk$, and semantics of $\rightadj\top\downarrow\functionA$)} \\ (\forall\numberk\lt\numbern_0) \functionA_\numberk & \text{ (definition of relative quantifier)} \\ \end{array}\) \(\bbox[10px, border:4px black solid] { \begin{array}{} & (\numbern_0\not\leq{?}) = ({?} \lt \numbern_0) \\ & \bbox[10px, border:2px black solid] { \begin{array}{} \numbern_0 \leq {?} \\ \bbox[10px, border:2px cyan dashed] {(\functionA\downarrow\leftadj\bot) = \setS} \\ \end{array} } \\ \N \\ \end{array} } \)
But (the last) is precisely (the antecedent for $PCON(\numbern_0)$),
so if [($\numbern_0$ is least in $\functionA\downarrow\leftadj\bot$) and ($PCON(\numbern_0)$ is valid)], then ($\functionA_{\numbern_0}$ is true).
Thus we have derived a contradiction from the joint assumptions that ($\numbern_0$ is least in $\functionA\downarrow\leftadj\bot$) and ($PCON(\numbern_0)$ is valid).
The rest of the argument is the same, viz: \[\begin{array}{} WO \wedge (\functionA\downarrow\bot \not= \emptyset) \wedge PCON \xRightarrow{WO} \big( (\exists\numbern_0\in\functionA\downarrow\bot)(\forall \numbern \in \functionA\downarrow\bot) (\numbern_0 \le \numbern) \big) \wedge PCON(\numbern_0) \xRightarrow{\text{by the above}} \bot \\ WO \wedge PCON \Longrightarrow \neg\neg(\functionA\downarrow\bot = \emptyset) \xRightarrow{\text{excluded middle}} (\functionA\downarrow\bot = \emptyset) \Leftrightarrow (\functionA = \top) \\ WO \Longrightarrow \big[ PCON \Rightarrow (\functionA = \top) \big] = PIND \\ \end{array}\]


$PIND \Rightarrow WO$
Suppose (a subset $\setS\subseteq\N$) has NO least element.
Then we claim the predicate $({?}\notin\setS) : \N \longrightarrow \bftwo : \numbern \mapsto (\numbern\notin\setS)$ satisfies the $PCON$.
Suppose ($\numbern\in\N$) is such that $(\forall\numberk\lt\numbern)(\numberk\notin\setS)$.
Then if ($\numbern\in\setS$), $\numbern$ is the least element of $\setS$. But $\setS$ has no least element. So ($\numbern\notin\setS$).
Thus the predicate satisfies $PCON$.
$PIND$ then implies $(\forall\numbern\in\N)(\numbern\notin\setS)$, i.e., $\setS$ is empty.
Taking the contrapositive, and applying (the law of the excluded middle), we have shown that $PIND$ implies:
$\N$ is well-ordered ($WO$):
$WO:{}$ Any (subset $\setS\subseteq\N$ which is non-empty) must have (a least element $\numbern_0$, i.e., $(\numbern_0\in\setS) \wedge (\forall\numbern\in\setS)(\numbern_0\leq\numbern)$).
The above argument may be expressed with more extensive use of logical notation: \[\begin{array}{} \neg(\text{$\setS$ has a least element}) \xRightarrow{\text{by the above}} PCON({?}\notin\setS) \xRightarrow{PIND({?}\notin\setS)} \big((\forall\numbern\in\N)(\numbern\notin\setS)\big) \Leftrightarrow (\setS=\emptyset) \\ PIND({?}\notin\setS) \wedge \neg(\text{$\setS$ has a least element}) \Longrightarrow (\setS=\emptyset) \\ PIND({?}\notin\setS) \Longrightarrow \Big[ \neg(\text{$\setS$ has a least element}) \Rightarrow (\setS=\emptyset) \Big] \xRightarrow{\text{contrapositive}} \Big[ (\setS\not=\emptyset) \Rightarrow \neg\neg(\text{$\setS$ has a least element}) \xLeftrightarrow{\text{excluded middle}} (\text{$\setS$ has a least element}) \Big] = WO(\setS) \\ \end{array}\]

$WO \Leftrightarrow PIND$

It is possible to give an “if and only if”-style proof of the equivalence of $WO$ and $PIND$.
First, we establish an “if and only if” relation between the “has a least element” and predecessors conditions.
The statements in the left column below, for an arbitrary (subset $\setS\subseteq\N$), and those in the right column below, for an arbitrary (predicate $\functionA : \N\to\bftwo$),
are logically equivalent.
(Here we use a briefer notation: unrestricted quantifiers are presumed to vary over all of $\N$.
The tautology referenced is between $A \Rightarrow B$ and $\neg(A \wedge \neg B)$, or a variant. The generic version of (the tautology used) is shown at the extreme left or right.
A $*$ or $\text{*}$ in a comment simply indicates the step is a key one.
At (the bottom of each column) is (a small 2-diagram in $\Cat$) showing (the logical equivalence) as (a 2-cell in $\Cat$), and giving (a name) to (that logical equivalence).
(The relations between these 2-cells are best explained in terms of the “mates calculus” in the 2-category $\Cat$.) )

\(\begin{array}{l} & \neg (\setS \text{ has a least element}) && \neg (\functionA\downarrow\bot \text{ has a least element}) \\ & \neg (\exists\numbern\in\setS) (\forall\numberk\in\setS) (\numbern\leq\numberk) & \text{(definition of least)} & \neg (\exists\numbern\in\functionA\downarrow\bot) (\forall\numberk\in\functionA\downarrow\bot) (\numbern\leq\numberk)\\ & (\forall\numbern\in\setS) (\exists\numberk\in\setS) (\numbern\not\leq\numberk) & \text{(negation and quantification)} \mkern20mu & (\forall\numbern\in\functionA\downarrow\bot) (\exists\numberk\in\functionA\downarrow\bot) (\numbern\not\leq\numberk)\\ & (\forall\numbern\in\setS) (\exists\numberk\in\setS) (\numberk\lt\numbern) & \text{(* $(\N,\leq)$ is a total order)} & (\forall\numbern\in\functionA\downarrow\bot) (\exists\numberk\in\functionA\downarrow\bot) (\numberk\lt\numbern) \\ & \forall\numbern \big[ \numbern\in\setS \Longrightarrow \exists\numberk(\numberk\lt\numbern \wedge \numberk\in\setS) \big] & \text{(definition of relative $\forall$, $\exists$)} & \forall\numbern \big[ \neg\functionA_\numbern \Longrightarrow \exists\numberk(\numberk\lt\numbern \wedge \neg\functionA_\numberk) \big] \\ & \forall\numbern \big[ \neg\exists\numberk(\numberk\lt\numbern \wedge \numberk\in\setS) \Longrightarrow \numbern\notin\setS \big] & \text{(* taking the contrapositive)} & \forall\numbern \big[ \neg\exists\numberk(\numberk\lt\numbern \wedge \neg\functionA_\numberk) \Longrightarrow \functionA_\numbern \big] \\ \neg(A \wedge B) & \forall\numbern \big[ \forall\numberk\neg(\numberk\lt\numbern \wedge \numberk\in\setS) \Longrightarrow \numbern\notin\setS \big] & \text{(negation and quantification)} & \forall\numbern \big[ \forall\numberk\neg(\numberk\lt\numbern \wedge \neg\functionA_\numberk) \Longrightarrow \functionA_\numbern \big] & \neg (A \wedge \neg B) \\ A \Rightarrow \neg B & \forall\numbern \big[ \forall\numberk [\numberk\lt\numbern \Rightarrow \numberk\notin\setS] \Longrightarrow \numbern\notin\setS \big] & \text{(* tautology for $\Rightarrow$)} & \forall\numbern \big[ \forall\numberk [\numberk\lt\numbern \Rightarrow \functionA_\numberk] \Longrightarrow \functionA_\numbern \big] & A \Rightarrow B \\ & \forall\numbern \big[ (\forall\numberk\lt\numbern)(\numberk\notin\setS) \Longrightarrow \numbern\notin\setS \big] & \text{(definition of relative $\forall$)} & \forall\numbern \big[ (\forall\numberk\lt\numbern)\functionA_\numberk \Longrightarrow \functionA_\numbern \big] \\ & PCON({?}\notin\setS) & \text{(definition of $PCON$)} & PCON(\functionA) \\ \\ & \begin{array}{} \llap{ \catI \xrightarrow{\textstyle \setS} {} \mkern8mu } \Sub (\N) & \xrightarrow{\textstyle \text{least}} & \bftwo^\op \\ & \llap{{?}\notin{?}'} \searrow & \Bigg\Updownarrow \rlap{\mkern-57mu \text{least}{+}{+}PCON} & \searrow \rlap\neg \\ && [\N,\bftwo] & \xrightarrow[\textstyle PCON]{} & \bftwo \\ \end{array} && \mkern-40mu \begin{array}{} && \Sub (\N) & \xrightarrow{\textstyle \text{least}} & \bftwo^\op \\ & \llap{{?''}\downarrow\bot} \nearrow && \Bigg\Updownarrow \rlap{\mkern-57mu \text{least}{-}{+}PCON} && \searrow \rlap\neg \\ \llap{ \catI \xrightarrow{\textstyle \functionA} {} \mkern8mu } [\N,\bftwo] &&& \xrightarrow[\textstyle PCON]{} &&& \bftwo \\ \end{array} \\ \end{array}\)

Then, for the (subset $\setS\subseteq\N$), we have:

\(\begin{array}{} PCON({?}\notin\setS) & \xleftrightarrow{\textstyle (\text{least}{+}{+}PCON)(\setS)} & \neg (\setS \text{ has a least element}) && (\setS \text{ has a least element}) \\ \llap{PIND({?}\notin\setS)} \Bigg\downarrow & \xLeftrightarrow{} & \Bigg\downarrow & \xLeftrightarrow[\textstyle \text{double negation}]{\textstyle \text{contrapositive plus}} & \Bigg\uparrow\rlap{WO(\setS)} \\ ({?}\notin\setS) = \top & \xleftrightarrow[\textstyle \text{definition of $\emptyset$}]{} & \setS=\emptyset && \setS\not=\emptyset \\ \end{array}\)
Thus, for every (subset $\setS\subseteq\N$), we have \(PIND({?}\notin\setS) \xLeftrightarrow{PIND{-}WO(\setS)} WO(\setS)\),
so if ($PIND$ is valid for all predicates on $\N$, e.g., ${?}\notin\setS$) then ($WO$ is valid for all subsets $\setS$ of $\N$).

To get the converse implication, the most efficient way is to use the bijection between (predicates $\functionA$ on $\N$) and (subsets $\setS$ of $\N$) described above.
For an arbitrary (predicate $\functionA:\N\to\bftwo$), let ($\setS = \functionA\downarrow\bot$), so that ($\functionA = {?}\notin\setS$).
Then we have $PIND(\functionA) = PIND({?}\notin\setS) \xLeftrightarrow{PIND{-}WO(\setS)} WO(\setS) = WO(\functionA\downarrow\bot)$,
so if ($WO$ is valid for all subsets of $\N$, e.g., $\setS = \functionA\downarrow\bot$) then ($PIND$ is valid for all predicates $\functionA$ on $\N$),
So $WO \Leftrightarrow PIND$.

An alternative (but equivalent) method is to redo the whole argument, replacing the (subset $\setS\subseteq\N$) with a (predicate $\functionA : \N \rightarrow \bftwo$).
For each (predicate $\functionA : \N \rightarrow \bftwo$) we use the logical equivalence ($\text{least}{-}{+}PCON$) at right above to get:

\(\begin{array}{} PCON(\functionA) & \xleftrightarrow{\textstyle (\text{least}{-}{+}PCON)(\functionA)} & \neg (\functionA\downarrow\bot \text{ has a least element}) && (\functionA\downarrow\bot \text{ has a least element}) \\ \llap{PIND(\functionA)} \Bigg\downarrow & \xLeftrightarrow{} & \Bigg\downarrow & \xLeftrightarrow[\textstyle \text{double negation}]{\textstyle \text{contrapositive plus}} & \Bigg\uparrow\rlap{WO(\functionA\downarrow\bot)} \\ \functionA = \top & \xleftrightarrow[]{} & \functionA\downarrow\bot = \emptyset && \functionA\downarrow\bot\not=\emptyset \\ \end{array}\)
Thus $PIND(\functionA) \Leftrightarrow WO(\functionA\downarrow\bot)$, and so on.

For comparison and reference, here are logical equivalents for the positive statements ($\setS$ has a least element) and ($\functionA\downarrow\bot$ has a least element).

\(\begin{array}{l} & (\setS \text{ has a least element}) && (\functionA\downarrow\bot \text{ has a least element}) \\ & (\exists\numbern\in\setS) (\forall\numberk\in\setS) (\numbern\leq\numberk) & \text{(definition of least)} & (\exists\numbern\in\functionA\downarrow\bot) (\forall\numberk\in\functionA\downarrow\bot) (\numbern\leq\numberk)\\ & (\exists\numbern\in\setS) (\forall\numberk\in\setS) (\numberk\not\lt\numbern) & \text{(* $(\N,\leq)$ is a total order)} & (\exists\numbern\in\functionA\downarrow\bot) (\forall\numberk\in\functionA\downarrow\bot) (\numberk\not\lt\numbern)\\ & (\exists\numbern\in\setS) \forall\numberk [\numberk\in\setS \Rightarrow \numberk\not\lt\numbern] & \text{(definition of relative $\forall$)} & (\exists\numbern\in\functionA\downarrow\bot) \forall\numberk [ \neg\functionA_\numberk \Rightarrow \numberk\not\lt\numbern] \\ & (\exists\numbern\in\setS) \forall\numberk [\numberk\lt\numbern \Rightarrow \numberk\notin\setS] & \text{(* taking the contrapositive)} & (\exists\numbern\in\functionA\downarrow\bot) \forall\numberk [ \numberk\lt\numbern \Rightarrow \functionA_\numberk ] \\ & (\exists\numbern\in\setS) (\forall\numberk\lt\numbern) (\numberk\notin\setS) & \text{(definition of relative $\forall$)} & (\exists\numbern\in\functionA\downarrow\bot) (\forall\numberk\lt\numbern) \functionA_\numberk \\ A \wedge B & \exists\numbern \big( (\forall\numberk\lt\numbern)(\numberk\notin\setS) \wedge \numbern\in\setS \big) & \text{(definition of relative $\exists$)} & \exists\numbern \big( (\forall\numberk\lt\numbern) \functionA_\numberk \wedge \neg\functionA_\numbern \big) & A \wedge \neg B \\ \neg [A \Rightarrow \neg B] & \exists\numbern \neg\big[ (\forall\numberk\lt\numbern)(\numberk\notin\setS) \Longrightarrow \numbern\notin\setS \big] & \text{(* tautology for $\Longrightarrow $)} & \exists\numbern \neg\big[ (\forall\numberk\lt\numbern)\functionA_\numberk \Longrightarrow \functionA_\numbern \big] & \neg [A \Rightarrow B] \\ & \neg\forall\numbern \big[ (\forall\numberk\lt\numbern)(\numberk\notin\setS) \Longrightarrow \numbern\notin\setS \big] & \text{(negation and quantification)} \mkern20mu & \neg\forall\numbern \big[ (\forall\numberk\lt\numbern)\functionA_\numberk \Longrightarrow \functionA_\numbern \big] \\ & \neg PCON({?}\notin\setS) & \text{(definition of $PCON$)} & \neg PCON(\functionA) \\ \\ & \mkern-80mu \begin{array}{} \llap{ \catI \xrightarrow{\textstyle \setS} {} \mkern8mu } \Sub (\N) &&& \xrightarrow{\textstyle \text{least}} &&& \bftwo^\op \\ & \llap{{?}\notin{?}'} \searrow && \Bigg\Updownarrow \rlap{\mkern-57mu \text{least}{+}{-}PCON} && \nearrow \rlap\neg \\ && [\N,\bftwo] & \xrightarrow[\textstyle PCON]{} & \bftwo \\ \end{array} && \mkern-40mu \begin{array}{} && \Sub (\N) & \xrightarrow{\textstyle \text{least}} & \bftwo^\op \\ & \llap{{?''}\downarrow\bot} \nearrow & \Bigg\Updownarrow \rlap{\mkern-57mu \text{least}{-}{-}PCON} & \nearrow \rlap\neg \\ \llap{ \catI \xrightarrow{\textstyle \functionA} {} \mkern8mu } [\N,\bftwo] & \xrightarrow[\textstyle PCON]{} & \bftwo \\ \end{array} \end{array}\)



$WO \Leftrightarrow DCC$

The well-ordering condition, in turn, is equivalent to the

Descending chain condition ($DCC$):
Every descending chain $\numbern_0 \gt \numbern_1 \gt \numbern_2 \gt \cdots$ of natural numbers (i.e., in $\N$) terminates, i.e.,
there are no infinite descending chains of natural numbers.

$WO \Rightarrow DCC$:
Given a descending chain $\numbern_0 \gt \numbern_1 \gt \numbern_2 \gt \cdots$,
let $\setS = \{\numbern_0, \numbern_1, \numbern_2, \ldots \}$.
By $WO$, $\setS$ has a least element, $\numbern_{\text{least}}$, such that $\forall\numberi (\numbern_{\text{least}} \leq \numbern_\numberi)$.
So the chain is finite and terminates with $\numbern_{\text{least}}$.

$DCC \Rightarrow WO$:
Let $\setS \subseteq \N$ be non-empty, so there exists a number, say $\numbern_0$, in $\setS$.
If $\numbern_0$ is least in $\setS$, we are done.
If not, there exists $\numbern_1\in\setS$ such that $\numbern_1 \lt \numbern_0$.
Continue.
By $DCC$, the chain generated must terminate, with say, $\numbern_0 \gt \cdots \gt \numbern_{\numberk-1} \gt \numbern_\numberk$.
The termination condition means $\neg(\exists\numbern\in\setS)(\numbern \lt \numbern_\numberk)$,
for otherwise the chain could have continued with $\numbern$.
Hence $(\forall\numbern\in\setS) \neg (\numbern \lt \numbern_\numberk)$, hence $(\forall\numbern\in\setS) (\numbern_\numberk \leq \numbern)$.
Hence $\numbern_\numberk$ is a (the) least element in $\setS$.
So $DCC \Rightarrow WO$.

Finally, we claim $DCC$ is clearly true for $\N$,
for a descending chain starting at $\numbern\in\N$ can have at most $\numbern+1$ elements:
$\numbern \gt \numbern-1 \gt \numbern-2 \gt \cdots \gt 2 \gt 1 \gt 0$.




$PIND \Rightarrow SIND$

We switch some of the arrows from double arrows to single arrows in this section, because of the lack of diagonal double arrows.

For all $\numbern\gt0$ we have:

\(\begin{array}{} (\forall\numberk\lt\numbern)\functionA_\numberk && \xrightarrow{\textstyle \pi_{\numbern-1}} && \functionA_{\numbern-1} \\ & \llap{PCON(\numbern)} \searrow & \Longleftarrow & \swarrow \rlap{SCON(\numbern-1) } \\ && \functionA_\numbern \\ \end{array}\) i.e. \(\begin{array}{} (\forall\numberk\lt\numbern)\functionA_\numberk \xRightarrow{\textstyle \pi_{\numbern-1}} \functionA_{\numbern-1} \xRightarrow{\textstyle SCON(\numbern-1)} \functionA_\numbern \\ SCON(\numbern-1) \wedge (\forall\numberk\lt\numbern)\functionA_\numberk \Longrightarrow \functionA_\numbern \\ SCON(\numbern-1) \Longrightarrow \big[(\forall\numberk\lt\numbern)\functionA_\numberk \Rightarrow \functionA_\numbern \big] = PCON(\numbern) \\ \end{array}\)
so $(\forall\numbern\gt0)\big[PCON(\numbern) \leftarrow SCON(\numbern-1)\big]$.
Also, $PCON(0) \leftarrow ICON$.
Thus $PCON \longleftarrow ICON \wedge SCON$.
Then we have:
\(\begin{array}{} PCON && \xleftarrow{} && ICON \wedge SCON \\ & \llap{PIND} \searrow & \Longrightarrow & \swarrow \rlap{SIND} \\ && (\functionA = \top) \\ \end{array}\) i.e. \(\begin{array}{} ICON \wedge SCON \xRightarrow{\text{by the above}} PCON \xRightarrow{\textstyle PIND} (\functionA=\top) \\ PIND \wedge ICON \wedge SCON \Longrightarrow (\functionA=\top) \\ PIND \Longrightarrow \big[(ICON \wedge SCON) \Rightarrow (\functionA=\top) \big] = SIND \\ \end{array}\)



$SIND \Rightarrow PIND$

Finally, one may show that (successor induction $SIND$) implies (predecessors induction $PIND$).
This follows from contemplation of the diagram (for arbitrary $\numbern\geq0$):
\[\begin{array}{} (\forall\numberk\lt\numbern+1)\functionA_\numberk & \xlongequal{} & (\forall\numberk\leq\numbern)\functionA_\numberk & = & AP_\numbern \\ \llap{PCON(\numbern+1,\functionA)}\bigg\downarrow & \searrow & \bigg\uparrow \rlap{\pi_{0 .. \numbern}} && \bigg\downarrow\rlap{SCON(\numbern,AP)} \\ \functionA_{\numbern+1} & \xleftarrow[\textstyle \pi_{\numbern+1}]{} & (\forall\numberl\leq\numbern+1)\functionA_\numberl & = & AP_{\numbern+1} \\ \end{array}\] (The diagonal arrow) is induced by [(the right universal property) of (the two arrows out of its target)].
Given (the indicated equalities between the objects), we may also consider (the diagonal) as [(an arrow) between (the two objects on the right)].
With these as (its source and target), it becomes (the desired $SCON(\numbern,AP)$).
Thus $(\forall\numbern\geq0)\big[PCON(\numbern+1,\functionA) \Rightarrow SCON(\numbern,AP)\big]$.
Also, $PCON(0,\functionA) = ICON(\functionA) = \functionA_0 = AP_0 = ICON(AP)$.
Then we have successively \[\begin{array}{} PCON(\functionA) \xRightarrow{\text{by the above}} \big(ICON(AP) \wedge SCON(AP)\big) \xRightarrow{\textstyle SIND(AP)} (\forall\numbern)AP_\numbern \xRightarrow{\textstyle (\forall\numbern)(AP_\numbern\xRightarrow{\textstyle \pi_\numbern} \functionA_\numbern)} (\forall\numbern)\functionA_\numbern \\ SIND(AP) \wedge PCON(\functionA) \Longrightarrow (\forall\numbern)\functionA_\numbern \\ SIND(AP) \Longrightarrow \big[PCON(\functionA) \Rightarrow (\forall\numbern)\functionA_\numbern\big] = PIND(\functionA) \; . \\ \end{array}\] So if ($SIND$ holds for all predicates on $\N$), then ($PIND$ does also).


In summary, we have proved the following implications regarding $\N$, its predicates, and its subsets: \[\begin{array}{} && WO && \longrightarrow \atop \longleftarrow && DCC \\ & \swarrow && \searrow\nwarrow \\ SIND && \longrightarrow \atop \longleftarrow && PIND \mkern30mu . \\ \end{array}\] Thus $WO$, $SIND$, $PIND$ and $DCC$ are logically equivalent.


References:

Induction Tutorial, 2011-09-29, CS221, University of Chicago
Ascending chain condition, Wikipedia web page
A Well Ordering Is A Consistent Choice Function, by Tom Leinster, 2018-11-12

No comments:

Post a Comment