What a system cannot say — page 2

A derivation of ((p → q) ∧ (q → r)) → (p → r), and the term it is. A natural-deduction derivation of ((p → q) ∧ (q → r)) → (p → r) with a lambda term written under each formula. The whole derivation is the term λh₁. λh₂. (π₂ h₁) ((π₁ h₁) h₂), and at every node the type of the term is the formula the derivation proves there.

Logic

Every derivation is a term

Write a variable for each assumption, an abstraction where one is discharged, and an application where an implication is used, and a natural-deduction derivation becomes a term. The formula it proves is the term's type, and checking the one is checking the other. A detour in the proof — a lemma introduced and at once used — is a term that simplifies, and simplifying it is removing the detour.

A implies B, and r ∨ s sits between them. A = ((p ∨ q) ∧ (p → r)) ∧ (q → s) implies B = (r ∨ s) ∨ t. A resolution refutation of A with the negation of B yields the interpolant r ∨ s. It uses only the shared atoms r, s, follows from A and implies B, and lies between the strongest and weakest interpolants.

Logic

The sentence between a premise and its consequence

When one formula implies another, something sits between them written only in the words the two have in common: a sentence the first implies and that implies the second. A refutation of the first together with the denial of the second hands such a sentence over, and every possible one lies between a strongest and a weakest that can be computed outright.

Logic

A failed search is a proof

Search for an assignment by branching on variables and backing up whenever a clause turns false. If every branch fails, the tree the search leaves behind is itself a resolution refutation: write at each branch point the resolvent of the clauses below it, and the top of the tree is the empty clause. So every limit on short refutations is a limit on every such search — and the pigeonhole clauses, whose refutations are long, defeat them all.

4 axioms against 4 finite algebras. A table of candidate axioms against finite Heyting algebras built from small orders, marking which algebras validate which axiom at every valuation.

Logic

Not one step but a continuum

Classical logic is the constructive system plus one axiom, which makes it sound as though there are two logics and one gap. There are uncountably many logics in that gap, each one a class of algebras, and the smallest separations between them fit in five elements.

The smallest algebra that refutes each formula. A table of formulas against the smallest finite Heyting algebra refuting each, found by searching every order on a few points, with the formulas no such algebra refutes marked.

Logic

Refutable in something small

A formula that is not a theorem of the constructive system fails in some finite algebra, and the algebra can be found by search. That single property is what makes the propositional logic decidable — and the predicate version loses the property and the decidability with it.

Two graphs every count agrees on, and one question that does not. Two sixteen-point graphs drawn on a four-by-four grid, with one point marked and its six neighbours highlighted in each, the neighbours forming two triangles in one and a six-cycle in the other.

Logic

One gadget defeats every refinement

Colour refinement fails on two triangles against a hexagon; its two-dimensional version fixes that and fails on a pair of strongly regular graphs. For every k there are two graphs the k-dimensional version cannot separate, and they are built from one local piece whose only symmetry is a parity.

Who survives with two pebbles and who with three, over 4 rounds. A table of three pairs of graphs with, for each, whether the duplicating player survives a two-pebble game and a three-pebble game played to a fixed depth.

Logic

The boundary at three variables

Restrict a sentence to two variable names and it can still be arbitrarily long, because the names are reused. What it cannot be is deep: every satisfiable two-variable sentence has a small model, so asking whether one is satisfiable is a bounded search. Allow a third name and the question becomes undecidable.

The Alexander matrix of the trefoil. The trefoil with its 3 arcs numbered and its 3 crossings lettered, beside the 3 by 3 matrix they give. A minor of the matrix is the Alexander polynomial t − 1 + t⁻¹, whose value at −1 is the determinant 3.

Topology

A polynomial behind the colourings

The figure-eight knot and the cinquefoil both have determinant five, so they admit exactly the same colourings, and every counting argument treats them as one. Put a variable where the colouring rule has a two and the determinant becomes a polynomial — and the two knots come apart.

The order type of a nonstandard model of arithmetic. The ordinary numbers as a run of dots, followed by 9 galaxies — copies of the integers — at the positions of c/2 up to 2c, ordered densely like the rationals, with c² beyond. Infinitely many more galaxies lie between those drawn.

Logic

A number larger than every number

Ask for a number bigger than 0, bigger than 1, bigger than 2, and so on for ever. Every finite piece of that request is granted by an ordinary number, so compactness grants all of it at once — in a structure that satisfies every sentence true of the whole numbers and still contains something beyond all of them. Nothing in first-order logic can say 'and nothing else'.

A countable structure grown by adding witnesses. Stages of a structure built from 0 and 1 by adding sums, products, negatives and roots of quadratics: 2, 4, 12, 158 elements between −3 and 3.

Logic

A countable field that passes for the line

The real numbers are uncountable, and every first-order sentence about their addition, multiplication and order is also true of a countable field inside them — the real algebraic numbers. Löwenheim and Skolem showed this is no quirk of the reals: every theory with an infinite model has a countable one, including set theory, which then contains sets it calls uncountable.

Logic

Six sentences from two quantifiers

One relation, two variables, 'for every' and 'there is': there are eight ways to arrange them and six different sentences come out. Which of them imply which is a small, complete diagram, found by checking all 512 relations on three points — and the diagram crosses over in the middle, which is where every confusion about the order of quantifiers lives.

"There is an x" is a shadow: x² + ax + 1 = 0 has a solution exactly when |a| ≥ 2. A grid with a horizontal and x vertical, marking the cells the curve x squared plus a x plus one equals zero passes through; beneath it, a strip marking the columns that contain a mark, which are exactly those with a at least two in size.

Logic

A quantifier is a shadow

'There is an x such that …' asks whether a column of a grid contains a mark — which is the same as asking whether a shape casts a shadow on the axis below it. Over the real numbers every such shadow can be described without the quantifier, by polynomial inequalities: 'x² + ax + 1 = 0 has a solution' is just a² ≥ 4. Over the whole numbers the same kind of shadow can carve out the primes, and any set a computer can list.

A periodic point and a wandering one, both near 0.3, parted by step 4. The distance between the orbit of a periodic point and the orbit of a point from a dense orbit, both starting in the same small interval, plotted against the step until the wandering orbit nears the point farthest from the periodic one.

Dynamics

Sensitivity comes free

The standard definition of chaos asks for three things: an orbit that goes everywhere, periodic orbits everywhere, and sensitive dependence on the starting point. The third, the one the word chaos is usually taken to mean, turns out to follow from the other two. A periodic point and a wandering point that start side by side must eventually part, because the wanderer has to visit places the periodic orbit never goes.

Weighted votes: two functions with a cut, and parity without one. For three functions of three letters, the eight assignments placed on a line by a weighted count, with true assignments filled and the threshold marked where one exists.

Logic

A plane through the cube

Some truth functions are weighted votes: give each letter a weight, add the weights of the true letters, and say yes when the total passes a threshold. On the cube of assignments, such a function is a plane cutting the true corners from the false. Majority is one. Exclusive-or is not, and never can be — and of the 65,536 functions of four letters, only 1,882 are. The ones that are are exactly what a single artificial neuron can compute.

A witness that 97 is prime: 5 has order 96. A table of the checks Lucas's test makes on 97 with base 5: the power (n − 1)/q for each prime q dividing n − 1, none of which is 1.

Number

An order that proves a prime

Fermat's little theorem is a test that primes pass and composites mostly fail, and it can be fooled. Run backwards, it cannot. If some number a has order exactly n − 1 modulo n, then n is prime — because only a prime has n − 1 numbers to cycle through. Checking that takes the prime factors of n − 1, which need proofs of their own, and the proofs nest into a tree that anyone can check: Pratt's certificate, which shows every prime has a short proof of being one.

Twelve units of shortfall, on every solid with three faces at a corner. A bar for each of 8 polyhedra with three faces at every vertex, divided into each face's shortfall from six sides; every bar has total length twelve, and the hexagons contribute nothing.

Topology

Twelve pentagons, whatever the hexagons

A football has twelve pentagons and twenty hexagons. A molecule of sixty carbon atoms has the same pattern, a molecule of seventy has twelve pentagons and twenty-five hexagons, and a geodesic dome of any size has twelve places where the pattern of six breaks. None of this is a coincidence of design: Euler's formula, rearranged, says that faces meeting three at a corner must fall short of hexagons by exactly twelve in total, and the hexagons are free.

A smallest model of ∀x (Ax → ∃y (By ∧ ¬Cy)) ∧ ∃x (Ax ∧ Cx) ∧ ∀x (Bx → ¬Ax). Three overlapping circles with some regions shaded as empty and a single dot in each occupied region, forming a model of a sentence of monadic first-order logic.

Logic

One thing in each region is enough

Give first-order logic its full apparatus of nested quantifiers but only one-place predicates, and every question about truth is still settled by the regions of a diagram. A predicate cannot tell apart two things in the same region, so no model ever needs more than one thing per region — and with three predicates there are only 255 models to try.

Counting walks that never revisit a square. Dots for the ratio of successive counts of self-avoiding walks and for the n-th root of the count, against the number of steps, both approaching a dashed horizontal line at the connective constant.

Probability

A walk that may not step where it has been

Forbid a walk on the square grid from ever revisiting a site and the number of possible n-step walks grows like 2.638ⁿ instead of 4ⁿ — a number nobody can write down exactly. On the honeycomb it is exactly √(2 + √2), proved in 2010. And the walks spread out like n to the three-quarters, faster than any ordinary walk, which physicists have used since 1949 and mathematicians still cannot prove.

Adjectives that describe themselves, and the one that cannot exist. A grid of 8 adjectives against the same adjectives as written words, marked where the property holds of the spelling; the diagonal marks self-describing adjectives, and a flipped diagonal row labelled heterological matches no row.

Logic

The word that cannot describe itself

Some adjectives describe themselves — 'short' is short — and some do not — 'lengthy' is not lengthy. Call the second kind heterological, and ask whether 'heterological' is heterological. It is exactly when it is not. The diagonal that beat every list of numbers, every set of sets and every provability predicate has been turned on words that describe words, and it proves that no language can contain a word for its own notion of describing, naming or truth.

Whole-number points in a strip, and the shadow they cast. The lattice points satisfying 2x ≤ 5y ≤ 2x + 1 for x from 0 to 30, and their projection onto the x-axis, which repeats every 5.

Logic

Arithmetic with addition alone

Over the real numbers, a quantifier's shadow is described by inequalities. Over the whole numbers with addition and multiplication, a shadow can be any set a computer can list. In between lies arithmetic with addition and no multiplication, and there the shadows are always the same kind of thing: a finite exception, then a pattern that repeats. The whole numbers made from coins worth 6, 9 and 20 are every number from 44 on; the squares, which need multiplication, never repeat at all.

Gluing two countermodels: p ∨ ¬p. Two small Kripke models, each refuting one half of a disjunction, and the model made by placing both above a new first stage, at which neither half is forced.

Logic

A proof that says which half

Classical logic proves p ∨ ¬p without any idea which half is true. The constructive system never does that: whenever it proves a disjunction, it proves one of the two halves. The reason is a picture — two countermodels placed side by side above a new first stage that forces neither — and the logics between the two lose the property exactly when their pictures are not allowed to be glued.

Which chains of truth values validate Gödel's disjunctions. A grid of Gödel's pigeonhole formulas in m letters against chains of n truth values, marking which chains validate which formula, forming a staircase where m exceeds n.

Logic

No table of truth values is enough

Two truth values decide classical logic. Gödel asked in 1932 whether some longer list of values could decide the constructive system, and answered with a pigeonhole: with n values, some two of n + 1 statements must share one, so a formula saying exactly that holds in every n-valued table and is not a theorem. The chains of truth values then descend forever, and what they share is a logic of its own — the logic of the real interval.

Tseitin's clauses on the cube. the cube with a variable on each edge and a charge of 0 or 1 at each vertex, exactly one vertex charged 1. The parity demands give 32 clauses that cannot all be true.

Logic

A contradiction that is only a sum

Put a variable on every edge of a graph and ask each vertex for an odd or an even number of true edges, with the demands adding up to odd. Add all the demands and every edge is counted twice, so the left side is zero and the right side is one: the contradiction is a single sum. Resolution cannot add. It has to reach the same conclusion clause by clause, and on a graph where every group of vertices has many edges leaving it, that takes exponentially long.

Unifying f(x, g(x)) with f(h(y), g(z)). Three term trees: f(x, g(x)), f(h(y), g(z)), and their common instance f(h(y), g(h(y))) under the most general unifier x ↦ h(y), z ↦ h(y).

Logic

Two terms made equal, and no more

Resolution with variables needs two literals to clash, and they clash only after something has been substituted for their variables. There are infinitely many substitutions that would do. One of them is the most general — every other is it followed by something more — and an algorithm of four rewriting rules finds it or proves there is none. That single computation turns the search for instances from guessing into arithmetic.