AFTD Auto-Formalizing Theoretical Domains

relu_nonneg

theoremverified

For any real number x, relu(x) is greater than or equal to zero.

Statement

theorem relu_nonneg (x : ℝ) : 0 ≤ relu x
Source
folklore
Verified
20 Oct 2026
Axioms
Classical.choiceQuot.soundpropext
Built on 1
  • reluThe rectified linear unit (ReLU) activation function maps each real number x to max(x, 0).

Read back from the Lean

For every real number $x$, $0 \le \operatorname{relu}(x)$.

Written by a model that saw only the Lean, never the English above. If the two disagree, that is worth a challenge.

The proof, explained

By definition, relu(x) is max(x, 0). The claim 0 ≤ relu(x) follows directly by applying the property that the second argument to max is always bounded above by the maximum, instantiated with x and 0.

  1. 0 ≤ relu x

    By definition, relu(x) is max(x, 0). Applying the standard inequality le_max_right with a = x and b = 0 directly gives 0 ≤ max(x, 0), which is 0 ≤ relu(x).

    line 2

    uses le_max_right

    theorem relu_nonneg (x : ℝ) : 0 ≤ relu x := le_max_right x 0
le_max_right
For any two elements a and b in a linearly ordered type, b ≤ max a b.

Written by a model (gemini-3.8-flash) from the Lean source alone (no goal states were available). Lean verified the proof; this explanation is a reading aid and is not itself machine-checked.

Lean source view module on GitHub

/-- The ReLU activation function is always non-negative. -/theorem relu_nonneg (x : ℝ) : 0 ≤ relu x := le_max_right x 0
See it in the graphDiscuss this resultChallenge it