Lean.Elab.Tactic.Omega.Core

Our internal representation of an argument "justifying" that a constraint holds on some coefficients. We'll use this to construct the proof term once a contradiction is found.

assumption (s : Omega.Constraint) (x : Omega.Coeffs) (i : Nat) : Justification s x
Problem.assumptions[i] generates a proof that s.sat' coeffs atoms
tidy {s : Omega.Constraint} {c : Omega.Coeffs} (j : Justification s c) : Justification (Omega.tidyConstraint s c) (Omega.tidyCoeffs s c)
The result of tidy on another Justification.
combine {s t : Omega.Constraint} {c : Omega.Coeffs} (j : Justification s c) (k : Justification t c) : Justification (s.combine t) c
The result of combine on two Justifications.
combo {s t : Omega.Constraint} {x y : Omega.Coeffs} (a : Int) (j : Justification s x) (b : Int) (k : Justification t y) : Justification (Omega.Constraint.combo a s b t) (Omega.Coeffs.combo a x b y)
A linear combo of two Justifications.
bmod (m : Nat) (r : Int) (i : Nat) {x : Omega.Coeffs} (j : Justification (Omega.Constraint.exact r) x) : Justification (Omega.Constraint.exact (r.bmod m)) (Omega.bmod_coeffs m i x)
The justification for the constraint constructed using "balanced mod" while eliminating an equality.

Instances For

Lean.Elab.Tactic.Omega.Justification.instToString

source

def Lean.Elab.Tactic.Omega.Justification.tidy? {s : Omega.Constraint} {c : Omega.Coeffs} (j : Justification s c) :

Option ((s' : Omega.Constraint) × (c' : Omega.Coeffs) × Justification s' c')

Wrapping for Justification.tidy when tidy? is nonempty.

Equations

j.tidy? = match Lean.Omega.tidy? (s, c) with | some val => some ⟨Lean.Omega.tidyConstraint s c, ⟨Lean.Omega.tidyCoeffs s c, j.tidy⟩⟩ | none => none

source

def Lean.Elab.Tactic.Omega.Justification.toString {s : Omega.Constraint} {x : Omega.Coeffs} :

Justification s x → String

Print a Justification as an indented tree structure.

Equations

One or more equations did not get rendered due to their size.
(Lean.Elab.Tactic.Omega.Justification.assumption s x i).toString = toString "" ++ toString x ++ toString " ∈ " ++ toString s ++ toString ": assumption " ++ toString i ++ toString ""

source

instance Lean.Elab.Tactic.Omega.Justification.instToString {s : Omega.Constraint} {x : Omega.Coeffs} :

ToString (Justification s x)

Equations

Lean.Elab.Tactic.Omega.Justification.instToString = { toString := Lean.Elab.Tactic.Omega.Justification.toString }

source

def Lean.Elab.Tactic.Omega.Justification.tidyProof (s : Omega.Constraint) (x : Omega.Coeffs) (v prf : Expr) :

Expr

Construct the proof term associated to a tidy step.

Equations

Lean.Elab.Tactic.Omega.Justification.tidyProof s x v prf = Lean.mkApp4 (Lean.Expr.const `Lean.Omega.tidy_sat []) (Lean.toExpr s) (Lean.toExpr x) v prf

source

def Lean.Elab.Tactic.Omega.Justification.combineProof (s t : Omega.Constraint) (x : Omega.Coeffs) (v ps pt : Expr) :

Expr

Construct the proof term associated to a combine step.

Equations

Lean.Elab.Tactic.Omega.Justification.combineProof s t x v ps pt = Lean.mkApp6 (Lean.Expr.const `Lean.Omega.Constraint.combine_sat' []) (Lean.toExpr s) (Lean.toExpr t) (Lean.toExpr x) v ps pt

source

def Lean.Elab.Tactic.Omega.Justification.comboProof (s t : Omega.Constraint) (a : Int) (x : Omega.Coeffs) (b : Int) (y : Omega.Coeffs) (v px py : Expr) :

Expr

Construct the proof term associated to a combo step.

Equations

One or more equations did not get rendered due to their size.

source

def Lean.Elab.Tactic.Omega.Justification.bmodProof (m : Nat) (r : Int) (i : Nat) (x : Omega.Coeffs) (v w : Expr) :

MetaM Expr

Construct the proof term associated to a bmod step.

Equations

One or more equations did not get rendered due to their size.

source

def Lean.Elab.Tactic.Omega.Justification.proof {s : Omega.Constraint} {c : Omega.Coeffs} (v : Expr) (assumptions : Array Proof) :

Justification s c → Proof

Constructs a proof that s.sat' c v = true

Equations

One or more equations did not get rendered due to their size.
Lean.Elab.Tactic.Omega.Justification.proof v assumptions (Lean.Elab.Tactic.Omega.Justification.assumption s c i) = assumptions[i]!

source

structure Lean.Elab.Tactic.Omega.Fact :

Type

A Justification bundled together with its parameters.

coeffs : Omega.Coeffs
The coefficients of a constraint.
constraint : Omega.Constraint
The constraint.
justification : Justification self.constraint self.coeffs
The justification of a derived fact.

Instances For

Lean.Elab.Tactic.Omega.Fact.instToString

source

instance Lean.Elab.Tactic.Omega.Fact.instToString :

ToString Fact

Equations

Lean.Elab.Tactic.Omega.Fact.instToString = { toString := fun (f : Lean.Elab.Tactic.Omega.Fact) => toString f.justification }

source

def Lean.Elab.Tactic.Omega.Fact.tidy (f : Fact) :

Fact

tidy, implemented on Fact.

Equations

f.tidy = match f.justification.tidy? with | some ⟨constraint, ⟨coeffs, justification⟩⟩ => { coeffs := coeffs, constraint := constraint, justification := justification } | none => f

source

def Lean.Elab.Tactic.Omega.Fact.combo (a : Int) (f : Fact) (b : Int) (g : Fact) :

Fact

combo, implemented on Fact.

Equations

One or more equations did not get rendered due to their size.

source

structure Lean.Elab.Tactic.Omega.Problem :

Type

A omega problem.

This is a hybrid structure: it contains both Expr objects giving proofs of the "ground" assumptions (or rather continuations which will produce the proofs when needed) and internal representations of the linear constraints that we manipulate in the algorithm.

While the algorithm is running we do not synthesize any new Expr proofs: proof extraction happens only once we've found a contradiction.

assumptions : Array Proof
The ground assumptions that the algorithm starts from.
numVars : Nat
The number of variables in the problem.
constraints : Std.HashMap Omega.Coeffs Fact
The current constraints, indexed by their coefficients.
equalities : Std.HashSet Omega.Coeffs
The coefficients for which constraints contains an exact constraint (i.e. an equality).
eliminations : List (Fact × Nat × Int)
Equations that have already been used to eliminate variables, along with the variable which was removed, and its coefficient (either 1 or -1). The earlier elements are more recent, so if these are being reapplied it is essential to use List.foldr.
possible : Bool
Whether the problem is possible.
proveFalse? : Option Proof
If the problem is impossible, then proveFalse? will contain a proof of False.
proveFalse?_spec : (self.possible || self.proveFalse?.isSome) = true
Invariant between possible and proveFalse?.
explanation? : Thunk String
If we have found a contradiction, explanation? will contain a human readable account of the deriviation.

Instances For

Lean.Elab.Tactic.Omega.Problem.instToString

source

def Lean.Elab.Tactic.Omega.Problem.isEmpty (p : Problem) :

Bool

Check if a problem has no constraints.

Equations

p.isEmpty = p.constraints.isEmpty

source

instance Lean.Elab.Tactic.Omega.Problem.instToString :

ToString Problem

Equations

One or more equations did not get rendered due to their size.

source

def Lean.Elab.Tactic.Omega.Problem.proveFalse {s : Omega.Constraint} {x : Omega.Coeffs} (j : Justification s x) (assumptions : Array Proof) :

Proof

Takes a proof that s.sat' x v for some s such that s.isImpossible, and constructs a proof of False.

Equations

One or more equations did not get rendered due to their size.

source

def Lean.Elab.Tactic.Omega.Problem.insertConstraint (p : Problem) :

Fact → Problem

Insert a constraint into the problem, without checking if there is already a constraint for these coefficients.

Equations

One or more equations did not get rendered due to their size.

source

def Lean.Elab.Tactic.Omega.Problem.addConstraint (p : Problem) :

Fact → Problem

Add a constraint into the problem, combining it with any existing constraints for the same coefficients.

Equations

One or more equations did not get rendered due to their size.

source

def Lean.Elab.Tactic.Omega.Problem.selectEquality (p : Problem) :

Option (Omega.Coeffs × Nat)

Walk through the equalities, finding either the first equality with minimal coefficient ±1, or otherwise the equality with minimal (r.minNatAbs, r.maxNatAbs) (ordered lexicographically).

Returns the coefficients of the equality, along with the value of minNatAbs.

Although we don't need to run a termination proof here, it's nevertheless important that we use this ordering so the algorithm terminates in practice!

Equations

One or more equations did not get rendered due to their size.

source

def Lean.Elab.Tactic.Omega.Problem.replayEliminations (p : Problem) (f : Fact) :

Fact

If we have already solved some equalities, apply those to some new Fact.

Equations

One or more equations did not get rendered due to their size.

source

def Lean.Elab.Tactic.Omega.Problem.solveEasyEquality (p : Problem) (c : Omega.Coeffs) :

Problem

Solve an "easy" equality, i.e. one with a coefficient that is ±1.

After solving, the variable will have been eliminated from all constraints.

Equations

One or more equations did not get rendered due to their size.

source

def Lean.Elab.Tactic.Omega.Problem.dealWithHardEquality (p : Problem) (c : Omega.Coeffs) :

OmegaM Problem

We deal with a hard equality by introducing a new easy equality.

After solving the easy equality, the minimum lexicographic value of (c.minNatAbs, c.maxNatAbs) will have been reduced.

Equations

One or more equations did not get rendered due to their size.

source

def Lean.Elab.Tactic.Omega.Problem.solveEquality (p : Problem) (c : Omega.Coeffs) (m : Nat) :

OmegaM Problem

Solve an equality, by deciding whether it is easy (has a ±1 coefficient) or hard, and delegating to the appropriate function.

Equations

p.solveEquality c m = if m = 1 then pure (p.solveEasyEquality c) else p.dealWithHardEquality c

source

partial def Lean.Elab.Tactic.Omega.Problem.solveEqualities (p : Problem) :

OmegaM Problem

Recursively solve all equalities.

source

def Lean.Elab.Tactic.Omega.Problem.addInequality_proof (c : Int) (x : Omega.Coeffs) (p : Proof) :

Proof

Constructing the proof term for addInequality.

Equations

One or more equations did not get rendered due to their size.

source

def Lean.Elab.Tactic.Omega.Problem.addEquality_proof (c : Int) (x : Omega.Coeffs) (p : Proof) :

Proof

Constructing the proof term for addEquality.

Equations

One or more equations did not get rendered due to their size.

source

def Lean.Elab.Tactic.Omega.Problem.addInequality (p : Problem) (const : Int) (coeffs : Omega.Coeffs) (prf? : Option Proof) :

Problem

Helper function for adding an inequality of the form const + Coeffs.dot coeffs atoms ≥ 0 to a Problem.

(This is only used while initializing a Problem. During elimination we use addConstraint.)

Equations

One or more equations did not get rendered due to their size.

source

def Lean.Elab.Tactic.Omega.Problem.addEquality (p : Problem) (const : Int) (coeffs : Omega.Coeffs) (prf? : Option Proof) :

Problem

Helper function for adding an equality of the form const + Coeffs.dot coeffs atoms = 0 to a Problem.

(This is only used while initializing a Problem. During elimination we use addConstraint.)

Equations

One or more equations did not get rendered due to their size.

source

def Lean.Elab.Tactic.Omega.Problem.addInequalities (p : Problem) (ineqs : List (Int × Omega.Coeffs × Option Proof)) :

Problem

Folding addInequality over a list.

Equations

One or more equations did not get rendered due to their size.

source

def Lean.Elab.Tactic.Omega.Problem.addEqualities (p : Problem) (eqs : List (Int × Omega.Coeffs × Option Proof)) :

Problem

Folding addEquality over a list.

Equations

One or more equations did not get rendered due to their size.

source

structure Lean.Elab.Tactic.Omega.Problem.FourierMotzkinData :

Type

Representation of the data required to run Fourier-Motzkin elimination on a variable.

var : Nat
Which variable is being eliminated.
irrelevant : List Fact
The "irrelevant" facts which do not involve the target variable.
lowerBounds : List (Fact × Int)
The facts which give a lower bound on the target variable, and the coefficient of the target variable in each.
upperBounds : List (Fact × Int)
The facts which give an upper bound on the target variable, and the coefficient of the target variable in each.
lowerExact : Bool
Whether the elimination would be exact, because all of the lower bound coefficients are ±1.
upperExact : Bool
Whether the elimination would be exact, because all of the upper bound coefficients are ±1.

Instances For

source

instance Lean.Elab.Tactic.Omega.Problem.instInhabitedFourierMotzkinData :

Inhabited FourierMotzkinData

Equations

One or more equations did not get rendered due to their size.

source

instance Lean.Elab.Tactic.Omega.Problem.instToStringFourierMotzkinData :

ToString FourierMotzkinData

Equations

One or more equations did not get rendered due to their size.

source

def Lean.Elab.Tactic.Omega.Problem.FourierMotzkinData.isEmpty (d : FourierMotzkinData) :

Bool

Is a Fourier-Motzkin elimination empty (i.e. there are no relevant constraints).

Equations

d.isEmpty = (d.lowerBounds.isEmpty && d.upperBounds.isEmpty)

source

def Lean.Elab.Tactic.Omega.Problem.FourierMotzkinData.size (d : FourierMotzkinData) :

Nat

The number of new constraints that would be introduced by Fourier-Motzkin elimination.

Equations

d.size = d.lowerBounds.length * d.upperBounds.length

source

def Lean.Elab.Tactic.Omega.Problem.FourierMotzkinData.exact (d : FourierMotzkinData) :

Bool

Is the Fourier-Motzkin elimination known to be exact?

Equations

d.exact = (d.lowerExact || d.upperExact)

source

def Lean.Elab.Tactic.Omega.Problem.fourierMotzkinData (p : Problem) :

Array FourierMotzkinData

Prepare the Fourier-Motzkin elimination data for each variable.

Equations

One or more equations did not get rendered due to their size.

source

def Lean.Elab.Tactic.Omega.Problem.fourierMotzkinSelect (data : Array FourierMotzkinData) :

MetaM FourierMotzkinData

Decides which variable to run Fourier-Motzkin elimination on, and returns the associated data.

We prefer eliminations which introduce no new inequalities, or otherwise exact eliminations, and break ties by the number of new inequalities introduced.

Equations

One or more equations did not get rendered due to their size.

source

def Lean.Elab.Tactic.Omega.Problem.fourierMotzkin (p : Problem) :

MetaM Problem

Run Fourier-Motzkin elimination on one variable.

Equations

One or more equations did not get rendered due to their size.

source

partial def Lean.Elab.Tactic.Omega.Problem.runOmega (p : Problem) :

OmegaM Problem

Run the omega algorithm (for now without dark and grey shadows!) on a problem.

source

partial def Lean.Elab.Tactic.Omega.Problem.elimination (p : Problem) :

OmegaM Problem

As for runOmega, but assuming the first round of solving equalities has already happened.