Monotonicity tactic #

The tactic mono applies monotonicity rules (collected through the library by being tagged @[mono]).

The version of the tactic here is a cheap partial port of the mono tactic from Lean 3, which had many more options and features. It is implemented as a wrapper on top of solve_by_elim.

Temporary syntax change: Lean 3 mono applied a single monotonicity rule, then applied local hypotheses and the rfl tactic as many times as it could. This is hard to implement on top of solve_by_elim because the counting system used in the maxDepth field of its configuration would count these as separate steps, throwing off the count in the desired configuration maxDepth := 1. So instead we just implement a version of mono in which monotonicity rules, local hypotheses and rfl are all applied repeatedly until nothing more is applicable. The syntax for this in Lean 3 was mono*. Both mono and mono* implement this behavior for now.

mono applies monotonicity rules and local hypotheses repetitively. For example,

example (x y z k : ℤ)
    (h : 3 ≤ (4 : ℤ))
    (h' : z ≤ y) :
    (k + 3 + x) - y ≤ (k + 4 + x) - z := by
  • One or more equations did not get rendered due to their size.
Instances For