tame/design/tpl/sec/notation.tex

643 lines
21 KiB
TeX
Raw Normal View History

% The TAME Programming Language Notational Conventions
%
% Copyright (C) 2021 Ryan Specialty, LLC.
%
% Licensed under the Creative Commons Attribution-ShareAlike 4.0
% International License.
%%
\section{Notational Conventions}
This section provides a fairly terse overview of the foundational
mathematical concepts used in this paper.
While we try to reason about \tame{} in terms of algebra,
first-order logic;
and set theory;
notation varies even within those branches.
To avoid ambiguity,
especially while introducing our own notation,
core operators and concepts are explicitly defined below.
This section begins its numbering at~0.
This is not only a hint that \tame{} (and this paper) use 0-indexing,
but also because equations; definitions; theorems; corollaries; and the
like are all numbered relative to their section.
When you see any of these prefixed with ``0.'',
this sets those references aside as foundational mathematical concepts
that are not part of the theory and operation of \tame{} itself.
\subsection{Propositional Logic}
\index{logic!propositional}
We reproduce here certain axioms and corollaries of propositional logic for
convenience and to clarify our interpretation of certain concepts.
The use of the symbols $\land$, $\lor$, and~$\neg$ are standard.
\indexsym\infer{infer}
\index{infer (\ensuremath\infer)}
The symbol $\infer$ means ``infer''.
We use $\implies$ in place of $\rightarrow$ for implication,
since the latter is used to denote the mapping of a domain to a codomain
in reference to functions.
We further use $\equiv$ in place of $\leftrightarrow$ to represent material
equivalence.
\indexsym\land{conjunction}
\index{conjunction (\ensuremath{\land})}
\begin{definition}[Logical Conjunction]
$p,q \infer (p\land q)$.
\end{definition}
\indexsym\lor{disjunction}
\index{disjunction (\ensuremath{\lor})}
\begin{definition}[Logical Disjunction]
$p \infer (p\lor q)$ and $q \infer (p\lor q)$.
\end{definition}
\begin{definition}[$\land$-Associativity]\dfnlabel{conj-assoc}
$(p \land (q \land r)) \infer ((p \land q) \land r)$.
\end{definition}
\begin{definition}[$\lor$-Associativity]\dfnlabel{disj-assoc}
$(p \lor (q \lor r)) \infer ((p \lor q) \lor r)$.
\end{definition}
\begin{definition}[$\land$-Commutativity]\dfnlabel{conj-commut}
$(p \land q) \infer (q \land p)$.
\end{definition}
\begin{definition}[$\lor$-Commutativity]\dfnlabel{disj-commut}
$(p \lor q) \infer (q \lor p)$.
\end{definition}
\begin{definition}[$\land$-Simplification]\dfnlabel{conj-simpl}
$p \land q \infer p$.
\end{definition}
\begin{definition}[Double Negation]\dfnlabel{double-neg}
$\neg\neg p \infer p$.
\end{definition}
\indexsym\neg{negation}
\index{negation (\ensuremath{\neg})}
\index{law of excluded middle}
\begin{definition}[Law of Excluded Middle]
$\infer (p \lor \neg p)$.
\end{definition}
\index{law of non-contradiction}
\begin{definition}[Law of Non-Contradiction]
$\infer \neg(p \land \neg p)$.
\end{definition}
\index{De Morgan's theorem}
\begin{definition}[De Morgan's Theorem]\dfnlabel{demorgan}
$\neg(p \land q) \infer (\neg p \lor \neg q)$
and $\neg(p \lor q) \infer (\neg p \land \neg q)$.
\end{definition}
\indexsym\equiv{equivalence}
\index{equivalence!material (\ensuremath{\equiv})}
\begin{definition}[Material Equivalence]
$p\equiv q \infer \big((p \land q) \lor (\neg p \land \neg q)\big)$.
\end{definition}
$\equiv$ denotes a logical identity.
Consequently,
it'll often be used as a definition operator.
\indexsym{\!\!\implies\!\!}{implication}
\index{implication (\ensuremath{\implies})}
\begin{definition}[Implication]\dfnlabel{implication}
$p\implies q \infer (\neg p \lor q)$.
\end{definition}
\begin{definition}[Tautologies]\dfnlabel{prop-taut}
$p\equiv (p\land p)$ and $p\equiv (p\lor p)$.
\end{definition}
\indexsym{\true}{boolean, true}
\indexsym{\false}{boolean, false}
\index{boolean!FALSE@\tamefalse{} (\false)}%
\index{boolean!TRUE@\tametrue{} (\true)}%
\begin{definition}[Truth Values]\dfnlabel{truth-values}
$\infer\true$ and $\infer\neg\false$.
\end{definition}
\indexsym\Int{integer}
\index{integer (\Int)}%
\begin{definition}[Boolean/Integer Equivalency]\dfnlabel{bool-int}
$\Set{0,1}\in\Int, \false \equiv 0$ and $\true \equiv 1$.
\end{definition}
\subsection{First-Order Logic and Set Theory}
\index{logic!first-order}
\indexsym\emptyset{set empty}
\indexsym{\Set{}}{set}
\index{set!empty (\ensuremath{\emptyset, \{\}})}
The symbol $\emptyset$ represents the empty set---%
the set of zero elements.
We assume that the axioms of ZFC~set theory hold,
but define $\in$ here for clarity.
\todo{Introduce set-builder notation, $\union$, $\intersect$.}
\indexsym\in{set membership}
\indexsym\union{set, union}
\indexsym\intersect{set, intersection}
\index{set!membership@membership (\ensuremath\in)}
\index{set!union (\ensuremath\union)}
\index{set!intersection (\ensuremath\intersect)}
\begin{definition}[Set Membership]
$x \in S \equiv \Set{x} \intersect S \not= \emptyset.$
\end{definition}
\index{domain of discourse}
$\forall$ denotes first-order universal quantification (``for all''),
and $\exists$ first-order existential quantification (``there exists''),
over some domain of discourse.
\indexsym\exists{quantification, existential}
\index{quantification!existential (\ensuremath\exists)}
\begin{definition}[Existential Quantification]\dfnlabel{exists}
$\Exists{x\in X}{P(x)} \equiv
\true \in \Set{P(x) \mid x\in X}$.
\end{definition}
\indexsym\forall{quantification, universal}
\index{quantification!universal (\ensuremath\forall)}
\begin{definition}[Universal Quantification]\dfnlabel{forall}
$\Forall{x\in X}{P(x)} \equiv \neg\Exists{x\in X}{\neg P(x)}$.
\end{definition}
\index{quantification!vacuous truth}
\begin{remark}[Vacuous Truth]\remlabel{vacuous-truth}
By \dfnref{exists}, $\Exists{x\in\emptyset}P \equiv \false$
and by \dfnref{forall}, $\Forall{x\in\emptyset}P \equiv \true$.
And so we also have the tautologies $\infer \neg\Exists{x\in\emptyset}P$
and $\infer \Forall{x\in\emptyset}P$.
Empty domains lead to undesirable consequences---%
in particular,
we must carefully guard against them in \dfnref{quant-conn} and
\dfnref{quant-elim} to maintain soundness.
\end{remark}
We also have this shorthand notation:
\index{quantification!\ensuremath{\forall x,y,z}}
\index{quantification!\ensuremath{\exists x,y,z}}
\begin{align}
\Forall{x,y,z\in S}P \equiv
\Forall{x\in S}{\Forall{y\in S}{\Forall{z\in S}P}}, \\
\Exists{x,y,z\in S}P \equiv
\Exists{x\in S}{\Exists{y\in S}{\Exists{z\in S}P}}.
\end{align}
\begin{definition}[Quantifiers Over Connectives]\dfnlabel{quant-conn}
Assuming that $x$ is not free in $\varphi$,
\begin{alignat*}{3}
\varphi\land\Exists{x\in X}{P(x)}
&\equiv \Exists{x\in X}{\varphi\land P(x)}, \\
\varphi\lor\Exists{x\in X}{P(x)}
&\equiv \Exists{x\in X}{\varphi\lor P(x)}
\qquad&&\text{assuming $X\neq\emptyset$}.
\end{alignat*}
\end{definition}
\begin{definition}[Quantifier Elimination]\dfnlabel{quant-elim}
$\Exists{x\in X}{\varphi} \equiv \varphi$ assuming $X\neq\emptyset$
and $x$ is not free in~$\varphi$.
\end{definition}
\subsection{Functions}
\indexsym{f, g}{function}
\indexsym\mapsto{function, map}
\indexsym\rightarrow{function, domain map}
\index{function}
\index{function!map (\ensuremath\mapsto)}
\index{map|see {function}}
\index{function!domain}
\index{function!codomain}
\index{domain|see {function, domain}}
\index{function!domain map (\ensuremath\rightarrow)}
The notation $f = x \mapsto x' : A\rightarrow B$ represents a function~$f$
that maps from~$x$ to~$x'$,
where $x\in A$ (the domain of~$f$) and $x'\in B$ (the co-domain of~$f$).
\indexsym\times{set, Cartesian product}
\index{set!Cartesian product (\ensuremath\times)}
A function $A\rightarrow B$ can be represented as the Cartesian
product of its domain and codomain, $A\times B$.
For example,
$x\mapsto x^2 : \Int\rightarrow\Int$ is represented by the set of ordered
pairs $\Set{(x,x^2) \mid x\in\Int}$, which looks something like
\begin{equation*}
\Set{\ldots,\,(0,0),\,(1,1),\,(2,4),\,(3,9),\,\ldots}.
\end{equation*}
\indexsym{[\,]}{function, image}
\index{function!image (\ensuremath{[\,]})}
\index{function!as a set}
The set of values over which some function~$f$ ranges is its \dfn{image},
which is a subset of its codomain.
In the example above,
both the domain and codomain are the set of integers~$\Int$,
but the image is $\Set{x^2 \mid x\in\Int}$,
which is clearly a subset of~$\Int$.
We therefore have
\begin{align}
A \rightarrow B &\subset A\times B, \\
f : A \rightarrow B &\infer f \subset A\times B, \\
f = \alpha \mapsto \alpha' : A \rightarrow B
&= \Set{(\alpha,\alpha')
\mid \alpha\in A \land \alpha'\in B}, \\
f[D\subseteq A] &= \Set{f(\alpha) \mid \alpha\in D} \subset B, \\
f[] &= f[A].
\end{align}
\indexsym{()}{tuple}
\index{tuple (\ensuremath{()})}
\index{relation|see {function}}
An ordered pair $(x,y)$ is also called a \dfn{$2$-tuple}.
Generally,
an \dfn{$n$-tuple} is used to represent an $n$-ary function,
where by convention we have $(x)=x$.
So $f(x,y) = f((x,y)) = x+y$.
If we let $t=(x,y)$,
then we also have $f(x,y) = ft$,
which we'll sometimes write as a subscript~$f_t$ where disambiguation is
necessary and where parenthesis may add too much noise;
this notation is especially well-suited to indexes,
as in $f_1$.
Binary functions are often written using \dfn{infix} notation;
for example, we have $x+y$ rather than $+(x,y)$.
\begin{equation}
f_x = f(x) \in \Set{b \mid (x,b) \in f}
\end{equation}
\subsubsection{Binary Operations On Functions}
\indexsym{R}{relation}
Consider two unary functions $f$ and~$g$,
and a binary relation~$R$.
\indexsym{\bicomp{R}}{function, binary composition}
\index{function!binary composition (\ensuremath{\bicomp{R}})}
We introduce a notation~$\bicomp R$ to denote the composition of a binary
function with two unary functions.
\begin{align}
f &: A \rightarrow B \\
g &: A \rightarrow D \\
R &: B\times D \rightarrow F \\
f \bicomp{R} g &= \alpha \mapsto f_\alpha R g_\alpha : A \rightarrow F
\end{align}
\indexsym\circ{function, composition}
\index{function!composition (\ensuremath\circ)}
Note that $f$ and~$g$ must share the same domain~$A$.
In that sense,
this is the mapping of the operation~$R$ over the domain~$A$.
This is analogous to unary function composition~$f\circ g$.
\index{function!constant}
A scalar value~$x$ can be mapped onto some function~$f$ using a constant
function.
For example,
consider adding some number~$x$ to each element in the image of~$f$:
\begin{equation*}
f \bicomp+ (\_\mapsto x) = \alpha \mapsto f_\alpha + x.
\end{equation*}
\indexsym{\_}{variable, wildcard}
\index{variable!wildcard/hole (\ensuremath{\_})}
The symbol~$\_$ is used to denote a variable that matches anything but is
never referenced,
and is often referred to as a ``wildcard'' (since it matches anything)
or a ``hole'' (since its value goes nowhere).
Note that we consider the bracket notation for the image of a function
$(f:A\rightarrow B)[A]$ to itself be a binary function.
Given that, we have $f\bicomp{[]} = f\bicomp{[A]}$ for functions returning
functions (such as vectors of vectors in \secref{vec}).
\subsection{Monoids and Sequences}\seclabel{monoids}
\index{abstract algebra!monoid}
\index{monoid|see abstract algebra, monoid}
\begin{definition}[Monoid]\dfnlabel{monoid}
Let $S$ be some set. A \dfn{monoid} is a triple $\Monoid S\monoidop e$
with the axioms
\begin{align}
\monoidop &: S\times S \rightarrow S
\tag{Monoid Binary Closure} \\
\Forall{a,b,c\in S&}{
a\monoidop(b\monoidop c) = (a\monoidop b)\monoidop c)
}, \tag{Monoid Associativity} \\
\Exists{e\in S&}{\Forall{a\in S}{e\monoidop a = a\monoidop e = a}}.
\tag{Monoid Identity}\label{eq:monoid-identity}
\end{align}
\end{definition}
\index{abstract algebra}
\index{abstract algebra!semigroup}
Monoids originate from abstract algebra.
A monoid is a semigroup with an added identity element~$e$.
Only the identity element must be commutative,
but if the binary operation~$\monoidop$ is \emph{also} commutative,
then the monoid is a \dfn{commutative monoid}.\footnote{%
A commutative monoid is less frequently referred to as an
\dfn{abelian monoid},
related to the common term \dfn{abelian group}.}
Consider some sequence of operations
$x_0 \monoidops x_n \in S$.
Intuitively,
a monoid tells us how to combine that sequence into a single element
of~$S$.
When the sequence has one or zero elements,
we then use the identity element $e\in S$:
as $x_0 \monoidop e = x_0$ in the case of one element
or $e \monoidop e = e$ in the case of zero.
\indexsym\cdots{sequence}
\index{sequence}
\begin{definition}[Monoidic Sequence]\dfnlabel{monoid-seq}
Generally,
given some monoid $\Monoid S\monoidop e$ and a sequence $\Fam{x}jJ\in S$
where $n<|J|$,
we have
$x_0\monoidop x_1\monoidops x_{n-1}\monoidop x_n$
represent the successive binary operation on all indexed elements
of~$x$.
When it's clear from context that the index is increasing by a constant
of~$1$,
that notation is shortened to $x_0\monoidops x_n$ to save
space.
When $|J|=1$, then $n=0$ and we have the sequence $x_0$.
When $|J|=0$, then $n=-1$,
and no such sequence exists,
in which case we expand into the identity element~$e$.
\end{definition}
For example,
given the monoid~$\Monoid\Int+0$,
the sequence $1+2+\cdots+4+5$ can be shortened to
$1+\cdots+5$ and represents the arithmetic progression
$1+2+3+4+5=15$.
If $x=\Set{1,2,3,4,5}$,
$x_0+\cdots+x_n$ represents the same sequence.
If $x=\Set{1}$,
that sequence evaluates to $1=1$.
If $x=\Set{}$,
we have $0$.
\index{conjunction!monoid}
\begin{lemma}\lemlabel{monoid-land}
$\Monoid\Bool\land\true$ is a commutative monoid.
\end{lemma}
\begin{proof}
$\Monoid\Bool\land\true$ is associative by \dfnref{conj-assoc}
and commutative by \dfnref{conj-commut}.
The identity element is~$\true\in\Bool$ by \dfnref{conj-simpl}.
\end{proof}
\index{disjunction!monoid}
\begin{lemma}\lemlabel{monoid-lor}
$\Monoid\Bool\lor\false$ is a commutative monoid.
\end{lemma}
\begin{proof}
$\Monoid\Bool\lor\false$ is associative by \dfnref{disj-assoc}
and commutative by \dfnref{disj-commut}.
The identity $\false\in\Bool$ follows from
\begin{alignat*}{3}
\false \lor p &\equiv p \lor \false &&\text{by \dfnref{disj-commut}} \\
&\equiv \neg(\neg p \land \neg\false)\qquad
&&\text{by \dfnref{demorgan}} \\
&\equiv \neg(\neg p) &&\text{by \dfnref{conj-simpl}} \\
&\equiv p. &&\text{by \dfnref{double-neg}} \tag*\qedhere
\end{alignat*}
\end{proof}
\goodbreak% Fits well on its own page, if we're near a page boundary
\subsection{Vectors and Index Sets}\seclabel{vec}
\tame{} supports scalar, vector, and matrix values.
Unfortunately,
its implementation history leaves those concepts a bit tortured.
A vector is a sequence of values, defined as a function of
an index.
An \dfn{index~set} is a set that is used to index values from another set;
they are usually subscripts of another set.
A \dfn{family} is a set that is indexed by the same index set.
In this paper,
we assume that an index set represents a range of integer values from $0$
to some number.
\begin{definition}[Family and Index Set]
Let $S$ be a family indexed by index set~$J$.
Then,
\begin{align}
\Fam{S}jJ,\qquad J = \Set{0,1,\dots,\len{J}-1}\in\PSet\Int.
\end{align}
\end{definition}
\indexsym{\PSet{S}}{set, power set}
\index{set!power set (\ensuremath{\PSet{S}})}
$\PSet{S}$ denotes the \dfn{power set} of $S$---%
the set of all subsets of~$S$ including $\emptyset$ and $S$~itself.
% TODO: font changes in index, making langle unavailable
%\indexsym{\Vector{}}{vector}
\index{vector!definition (\ensuremath{\Vector{}})}
\index{sequence|see vector}
\indexsym\Vectors{vector}
\index{real number (\ensuremath\Real)}
\indexsym\Real{real number}
\indexsym{\Fam{a}jJ}{index set}
\index{family|see {index set}}
\index{index set!_@notation (\ensuremath{\Fam{a}jJ})}
\begin{definition}[Vector]\dfnlabel{vec}
Let $J\subset\Int$ represent an index set.
A \dfn{vector}~$v\in\Vectors^\Real$ is a totally ordered sequence of
elements represented as a function of an element of its index set:
\begin{equation}\label{vec}
v = \Vector{v_0,\ldots,v_j}^{\Real}_{j\in J}
= j \mapsto v_j : J \rightarrow \Real.
\end{equation}
\end{definition}
This definition means that $v_j = v(j)$,
making the subscript a notational convenience.
We may omit the superscript such that $\Vectors^\Real=\Vectors$
and $\Vector{\ldots}^\Real=\Vector{\ldots}$.
When appropriate,
a vector may also be styled in a manner similar to linear algebra,
noting that our indexes begin at $0$ instead of~$1$:
\begin{equation}
\Vector{v_0,\dots,v_j}^\Real_{j\in J} =
\begin{bmatrix}
v_0 \\
\vdots \\
v_j
\end{bmatrix}_{j\in J}
=
\begin{bmatrix}
v_0 \\
\vdots \\
v_j
\end{bmatrix}.
\end{equation}
\index{matrix@see {vector}}
\index{vector!matrix}
\begin{definition}[Matrix]\dfnlabel{matrix}
Let $J\subset\Int$ represent an index set.
A \dfn{matrix}~$M\in\Matrices$ is a totally ordered sequence of
elements represented as a function of an element of its index set:
\begin{equation}
M = \Vector{M_0,\ldots,M_j}^{\Vectors^\Real}_{j\in J}
= j \mapsto M_j : J \rightarrow \Vectors^\Real.
\end{equation}
\end{definition}
The consequences of \dfnref{matrix}---%
defining a matrix as a vector of independent vectors---%
are important.
This defines a matrix to be more like a multidimensional array,
with no requirement that the lengths of the vectors be equal.
\begin{corollary}[Matrix Row Length Variance]\corlabel{matrix-row-len}
$\infer \Exists{M\in\Matrices}{\neg\Forall*{j}{\Forall{k}{\len{M_j} = \len{M_k}}}}$.
\end{corollary}
\corref{matrix-row-len} can be read ``there exists some matrix~$M$ such that
not all row lengths of~$M$ are equal''.
In other words---%
the inner vectors of a matrix can vary in length.
However,
certain systems (such as that of \axmref{class-intro}) may place
restrictions by specifying the inner index set as a dependent type:
\begin{equation}
\MFam{M}jJkK : J \rightarrow K_j \rightarrow \Real, \quad K : J \rightarrow \PSet\Int.
\end{equation}
\index{vector!matrix!rectangular}
This makes $K$ a set of index sets.
When $\len{K[J]}=1$
(that is---all $K_j$ are the same index set),
the matrix is \dfn{rectangular},
and can be written in a manner similar to linear algebra,
noting that our indexes begin at $0$ instead of~$1$;
that we use double-subscripts
(since matrices are functions returning functions);
and that we use $j,k$ in place of~$m,n$.
\begin{equation}
\begin{bmatrix}
M_{0_0} & M_{0_1} & \dots & M_{0_k} \\
M_{1_0} & M_{0_1} & \dots & M_{0_k} \\
\vdots & \vdots & \ddots & \vdots \\
M_{j_0} & M_{j_1} & \dots & M_{j_k} \\
\end{bmatrix}_{\underset{k\in K_0}{j\in J}}
\qquad
\text{if $|K[J]|=1$}.
\end{equation}
We may optionally omit the domains as in the vector notation.
\indexsym\eleundef{undefined}
\index{undefined}
If a matrix is \emph{not} rectangular,
the symbol~$\eleundef$ can be used to explicitly denote that specific scalar
values are undefined;
this is useful when the matrix representation is desirable when
describing the transformation of non-rectangular data \emph{into}
rectangular data.
For example,
\begin{equation}
\begin{bmatrix}
0 & 1 & 2 \\
3 & 4 & \eleundef \\
5 & \eleundef & \eleundef
\end{bmatrix}_{\underset{k\in K_j}{j\in J}}
=
\Vector{\Vector{0,1,2},\Vector{3,4},\Vector{5}},
\qquad
\begin{aligned}
J &= \Set{0,1,2}, \\
K &= \Set{(0,\Set{0,1,2}),\,
(1,\Set{0,1}),\,
(2,\Set{0})}.
\end{aligned}
\end{equation}
% TODO: symbol does not render properly in index
\begin{definition}[Rank]\dfnlabel{rank}
\index{rank}
The \dfn{rank} of some variable~$x$ is an integer value
\begin{equation*}
2021-05-20 15:28:10 -04:00
\rank{x} =
\begin{cases}
2 &x\in\Matrices, \\
1 &x\in\Vectors^\Real, \\
0 &x\in\Real.
\end{cases}
\end{equation*}
\end{definition}
Intuitively, the rank represents the number of dimensions of some variable~$x$.
A scalar has zero dimensions (a point);
a vector has one (a line);
and a matrix has two (a plane).
\index{dimensions@see {rank}}
\index{rank!dimensions}
In \tame{},
the rank is referred to as \dfn{dimensions} using the attribute
\xmlattr{dim}.
\subsection{XML Notation}
\indexsym{\xml{<>}}{XML}
\index{XML!notation (\xml{<>})}
The grammar of \tame{} is XML.
Equivalence relations will be used to map source expressions to an
underlying mathematical expression.
For example,
\begin{equation*}
\xml{<foo bar="$x$" baz="$y$" />} \equiv x = y
\end{equation*}
\noindent
defines that pattern of \xmlnode{foo} expression to be materially
equivalent to~$x=y$---%
anywhere an equality relation appears,
you could equivalently replace it with that XML representation without
changing the meaning of the mathematical expression.
Variables may also bind to literals in an XML expression.
For example,
\begin{equation*}
\xml{<quux $\alpha$ />},\qquad\alpha\in\Set{\emptystr, \xml{bar="baz"}}
\end{equation*}
\noindent
can represent either \xml{<quux />} or \xml{<quux bar="baz" />}.
\index{empty string}
$\emptystr$~represents the empty string.
Any text typeset in \texttt{typewriter} represents a literal
string of characters.