A crash course in first-order logicTweet
This is a short but complete description of first-order logic, a rich system to reason about objects and their relationships. I care mostly about first-order logic for its role in statistical relational learning (a branch of machine learning), which I'll cover in another post. Here, I just describe logic. If you want a more detailed explanation, see Russell and Norvig's excellent A.I. book.
Terms represent objects and relationships between objects, they are the reason why logic is so flexible. There are three types of terms:
- Constants represent objects, e.g.: Tokyo, the number 47, Cylon, Lion.
- Variables range over objects, e.g. the variable c could represent a city, x an integer, s a species. By convention, variables start with a lowercase character.
- Functions are mappings between a list of objects to another object, e.g. CapitalOf could take a country and return a city, while Multiply takes two numbers and returns a number.
Add is a function taking two numbers and returning a number, x is a variable, and 5 is a constant. Since a function is a term, it can be used within functions:\[Add(Multiply(x, y), 5).\]
First-order logic formulas ultimately resolve to a truth value: True or False, yet the terms we've seen return objects. To get a truth value, we need predicates. A predicate is like a function but it maps terms to a truth value instead of mapping them to a term. For example, if we want to say that adding 0 to x yields x, we can write:\[Equals(Add(x, 0), x).\]
It's common to use the equal sign for the "equals" (or identity) predicate:\[Add(x, 0) = x.\]
Equals is a predicate. In this case it takes two numbers and returns true or false. We could have a predicate taking three cities and return true if they are on the same continent:\[SameContinent(Toronto, c, CapitalOf(LargestCountryOf(Europe))),\]
where SameContinent is a predicate, Toronto and Europe are constants, c a variable ranging over cities, and both CapitalOf and LargestCountryOf are functions taking a single argument.
An atomic sentence is something that, alone, is a valid first-order logic formula. A predicate is an atomic sentence since it returns a truth value, but we also have two special symbols: True and False.
True is also called top, and can be represented with the symbol \(\top\).
False is also called bottom, and can be represented with the symbol \(\bot\).
You can connect sentences with connectives to form complex sentences. The standard connectives are:
- The binary connective and: \(x \land y\), which is true only if both \(x\) and \(y\) are true. Like all other connectives shown here, if \(x\) and \(y\) are sentences, then \(x \land y\) is also a valid sentence.
- The binary connective or: \(x \lor y\), which is true only if \(x\) is true, if \(y\) is true, or if both are true.
- The binary connective implies: \(x \Rightarrow y\), returns true in all cases, except if \(x\) is true and \(y\) is false.
- The binary connective iff: \(x \iff y\), returns true if \(x\) and \(y\) have the same value, that is if they are both true, or both false.
- The binary connective xor (exclusive or): \(x \oplus y\), returns true if \(x\) and \(y\) have different values.
- The unary connective not: \(\lnot x\), which is true only if \(x\) is false.
Be careful with implication, there is nothing wrong with it, except that it doesn't fit how we use the term implies. For example: \(Equals(AgeOf(Earth), 42) \Rightarrow StillAlive(Elvis)\) is true. If you're confused, read again the description of implies.
There are two qualifiers in first-order logic: the universal qualifier "for all" denoted \(\forall\), and the existential qualifier "exists" denoted \(\exists\). They, well, qualify variables, e.g.:\[\forall x: Multiply(x, 0) = 0\]
reads "for all x, multiplying x by 0 yields 0". Normally, all variables should be qualified to determine their scope, e.g.:\[\forall x \exists y: RealNumber(x) \Rightarrow GreaterThan(y, x)\]
which means, "for all real numbers x, there is a number y that is greater than x".
A last example: we can express "there is a color c brighter than x, unless x is white" with:\[\forall x \exists c: IsWhite(x) \lor BrighterThan(c, x).\]