explicitApproximant_torusDistance
For any Boolean function f : (Fin n → Bool) → Bool on n > 0 variables and any x : Fin n → Bool, torusDistance (explicitApproximant f x - bit (f x) / 2) equals explicitError f.
Statement
theorem explicitApproximant_torusDistance {n : ℕ} (hn : 0 < n)
(f : (Fin n → Bool) → Bool) (x : Fin n → Bool) :
torusDistance (explicitApproximant f x - bit (f x) / 2) = explicitError f- Formalizes
- arXiv:2610.19174, Listing A.14
- Verified
- 21 Oct 2026
- Axioms
Classical.choiceQuot.soundpropext- Built on 8
- explicitApproximantFor a Boolean function f : (Fin n → Bool) → Bool and input x : Fin n → Bool, explicitApproximant f x is (if f x then 1 else 0)/2 -…
- torusDistanceThe distance on the one-dimensional torus R/Z, defined as the distance of a real number to the nearest integer.
- explicitErrorThe explicit torus approximation error: 2^{-n-1} if the number of accepting inputs of f is odd, and 0 if even.
- bitThe real embedding of a Boolean value, sending false to 0 and true to 1.
- parityCharacterFor an implicit natural number n and a Boolean assignment x : Fin n → Bool, parityCharacter x is the real number defined by the product…
- topCoefficientFor an implicit natural number n and a Boolean-valued function f : (Fin n → Bool) → Bool, topCoefficient f is the integer defined by the…
- acceptingCountThe number of accepting inputs |f⁻¹(1)| of a Boolean function f.
- fullMonomialThe full monomial x_[n] = ∏_{i=1}^n x_i evaluated at a point in {0,1}^n.
Read back from the Lean
For every natural number $n$ satisfying $0 < n$, every Boolean function $f : (\text{Fin } n \to \text{Bool}) \to \text{Bool}$, and every input vector $x : \text{Fin } n \to \text{Bool}$, the torus distance (distance to the nearest integer, $|u - \operatorname{round}(u)|$) of $\operatorname{explicitApproximant}(f, x) - \operatorname{bit}(f(x)) / 2$ equals $\operatorname{explicitError}(f)$. Here: - $\operatorname{bit}(b)$ is $1$ if $b = \text{true}$ and $0$ if $b = \text{false}$. - $\operatorname{explicitError}(f)$ is $1 / 2^{n+1}$ if the number of inputs satisfying $f$ (i.e., $\operatorname{acceptingCount}(f) = |\{y \mid f(y) = \text{true}\}|$) is odd, and $0$ if it is even. - $\operatorname{explicitApproximant}(f, x)$ is defined as $(1/2 \text{ if } f(x) \text{ else } 0) - \lfloor \operatorname{topCoefficient}(f) / 2 \rfloor \cdot \operatorname{fullMonomial}(x) - (\operatorname{topCoefficient}(f) \bmod 2) \cdot \operatorname{parityCharacter}(x) / 2^{n+1}$.
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
The explicit approximant differs from bit(f(x)) / 2 by an integer multiple of the full monomial plus a scaled parity character. Because the full monomial takes values in {0, 1}, its contribution is an integer shift that leaves the torus distance unchanged. Since the parity character has absolute value 1 and the denominator 2^(n + 1) is at least 4, the remaining term has absolute value at most 1/4 < 1/2, so its nearest integer is 0 and its torus distance is simply (topCoefficient f % 2) / 2^(n + 1). Finally, the parity of topCoefficient f matches that of acceptingCount f, which equals explicitError f in both the odd and even cases.
Every finite product of ±1 factors is 1 or -1, so the character product over all coordinates is 1 or -1.
We prove by induction on the finite set s that the product of (if x i = true then 1 else -1) over s is 1 or -1. The empty product is 1. When inserting a new element a, the previous product is multiplied by ±1; in all four combinations of signs, the product remains 1 or -1. Specialising to the entire universe Finset.univ yields the result.
uses
Finset.induction_on,Finset.prod_insert,Finset.univhave key : ∀ s : Finset (Fin n), (∏ i ∈ s, if x i = true then (1:ℤ) else -1) = 1 ∨ (∏ i ∈ s, if x i = true then (1:ℤ) else -1) = -1 := by intro s induction s using Finset.induction_on with | empty => simp | insert a s ha ih => rw [Finset.prod_insert ha] rcases ih with h | h <;> rcases (show (if x a = true then (1:ℤ) else -1) = 1 ∨ (if x a = true then (1:ℤ) else -1) = -1 from by split <;> simp) with h' | h' <;> rw [h', h] <;> norm_num simpa using key Finset.univ
The top Fourier coefficient topCoefficient f and the accepting count acceptingCount f have the same parity.
We first express the accepting count as the sum of indicator booleans ∑ x, (if f x = true then 1 else 0). Subtracting this from topCoefficient f gives a single sum over x of (∏ i, χ i(x) - 1) for accepting inputs and 0 - 0 for rejecting inputs. By the previous lemma, the character product is ±1, so on accepting inputs the summand is either 1 - 1 = 0 or -1 - 1 = -2. Since each summand is divisible by 2, their sum is divisible by 2, which means topCoefficient f and acceptingCount f are congruent modulo 2.
uses
Finset.sum_boole,Fintype.card_subtype,Finset.sum_sub_distrib,Finset.dvd_sum,if_pos,explicitApproximant_torusDistance_chi_int_pm,if_neg,Int.emod_eq_emod_iff_emod_sub_eq_zero,Int.emod_eq_zero_of_dvdhave hsum : (∑ x : Fin n → Bool, (if f x = true then (1:ℤ) else 0)) = (acceptingCount f : ℤ) := by rw [Finset.sum_boole] rw [acceptingCount, Fintype.card_subtype] have hdvd : (2:ℤ) ∣ topCoefficient f - (acceptingCount f : ℤ) := by have hsplit : topCoefficient f - (acceptingCount f : ℤ) = ∑ x : Fin n → Bool, ((if f x = true then ∏ i, (if x i = true then (1:ℤ) else -1) else 0) - (if f x = true then (1:ℤ) else 0)) := by rw [Finset.sum_sub_distrib, hsum] rfl rw [hsplit] refine Finset.dvd_sum ?_ intro y _ by_cases h : f y = true · rw [if_pos h, if_pos h] rcases explicitApproximant_torusDistance_chi_int_pm y with hc | hc <;> rw [hc] <;> norm_num · rw [if_neg h, if_neg h] simp exact (Int.emod_eq_emod_iff_emod_sub_eq_zero).mpr (Int.emod_eq_zero_of_dvd hdvd)
The parity character parityCharacter x takes value 1 or -1 for every Boolean input x.
Unfolding the definition of parityCharacter, we perform induction on subsets s ⊆ Fin n to show that the real product of ±1 factors is 1 or -1. The base case is the empty product 1, and each extension multiplies an element of {1, -1} by ±1, preserving membership in {1, -1}. Evaluating on the full set of coordinates Finset.univ gives the claim.
uses
Finset.induction_on,Finset.prod_insert,Finset.univrw [parityCharacter] have key : ∀ s : Finset (Fin n), (∏ i ∈ s, (if x i = true then (1:ℝ) else -1)) = 1 ∨ (∏ i ∈ s, (if x i = true then (1:ℝ) else -1)) = -1 := by intro s induction s using Finset.induction_on with | empty => simp | insert a s ha ih => rw [Finset.prod_insert ha] rcases ih with h | h <;> rcases (show (if x a = true then (1:ℝ) else -1) = 1 ∨ (if x a = true then (1:ℝ) else -1) = -1 from by split <;> simp) with h' | h' <;> rw [h', h] <;> norm_num simpa using key Finset.univ
The full monomial fullMonomial x evaluates to 0 or 1 at any Boolean point x.
If x i = true for every coordinate i, then every bit factor is 1, so the full monomial product is 1. Otherwise, there is some coordinate i where x i ≠ true, which makes bit (x i) = 0 and forces the entire product to 0.
uses
Finset.prod_eq_one,Finset.prod_eq_zero,Finset.mem_univby_cases h : ∀ i, x i = true · right rw [fullMonomial] refine Finset.prod_eq_one ?_ intro i _ rw [h i] simp [bit] · left push Not at h obtain ⟨i, hi⟩ := h rw [fullMonomial] refine Finset.prod_eq_zero (Finset.mem_univ i) ?_ simp [bit, hi]
Shifting by an integer does not change the torus distance: torusDistance (u + m) = torusDistance u.
By definition, torusDistance is |u - round u|. Because rounding commutes with integer translation (round(u + m) = round u + m), the integer m cancels inside the absolute value, leaving |u - round u|.
uses
round_add_intCastunfold torusDistance rw [round_add_intCast] congr 1 push_cast ring
The difference explicitApproximant f x - bit (f x) / 2 cancels the leading term, leaving the monomial and character terms.
Unfolding explicitApproximant and bit, the first term is bit (f x) / 2. Subtracting bit (f x) / 2 cancels it by basic ring algebra, leaving -(topCoefficient f / 2 * fullMonomial x) - (topCoefficient f % 2 * parityCharacter x / 2^(n + 1)).
have hsub : explicitApproximant f x - bit (f x) / 2 = -(↑(topCoefficient f / 2) * fullMonomial x) - ↑(topCoefficient f % 2) * parityCharacter x / 2 ^ (n + 1) := by simp only [explicitApproximant, bit] ring rw [hsub]
The full monomial term is an integer, so the difference is an integer shift of the character term, preserving the torus distance.
Let r = topCoefficient f % 2 and k = topCoefficient f / 2. Since fullMonomial x is either 0 or 1, the term -k * fullMonomial x is an integer (either 0 or -k). By integer invariance of torusDistance, this integer shift can be dropped, leaving torusDistance (-r * parityCharacter x / 2^(n + 1)).
uses
explicitApproximant_torusDistance_fullMonomial_zero_or_one,explicitApproximant_torusDistance_add_intset r := topCoefficient f % 2 with hrdef set k := topCoefficient f / 2 with hkdef have hint : ∃ m : ℤ, (-(↑k * fullMonomial x) - ↑r * parityCharacter x / 2 ^ (n + 1)) = (-(↑r * parityCharacter x / 2 ^ (n + 1))) + (m : ℝ) := by rcases explicitApproximant_torusDistance_fullMonomial_zero_or_one x with h | h · refine ⟨0, ?_⟩ rw [h] push_cast ring · refine ⟨-k, ?_⟩ rw [h] push_cast ring obtain ⟨m, hm⟩ := hint rw [hm, explicitApproximant_torusDistance_add_int]
Since n > 0, the denominator 2^(n + 1) is at least 4 and strictly positive.
Because n ≥ 1, we have n + 1 ≥ 2. Monotonicity of powers gives 4 = 2^2 ≤ 2^(n + 1), and positivity holds automatically.
uses
pow_le_pow_right₀have hpow : (4:ℝ) ≤ 2 ^ (n + 1) := by have h2 : 2 ≤ n + 1 := by omega calc (4:ℝ) = 2 ^ 2 := by norm_num _ ≤ 2 ^ (n + 1) := pow_le_pow_right₀ (by norm_num) h2 have hpos : (0:ℝ) < 2 ^ (n + 1) := by positivity
The absolute value |-r * parityCharacter x / 2^(n + 1)| equals r / 2^(n + 1), which is strictly less than 1/2.
The remainder r modulo 2 is 0 or 1, so 0 ≤ r ≤ 1. The parity character has absolute value 1, so the absolute value of the expression simplifies to r / 2^(n + 1). Since r ≤ 1 and 2^(n + 1) ≥ 4, this quotient is at most 1/4, which is strictly less than 1/2.
uses
Int.emod_two_eq_zero_or_one,explicitApproximant_torusDistance_parityCharacter_pm,abs_neg,abs_div,abs_mul,abs_of_nonneg,abs_of_pos,div_lt_iff₀have hr01 : r = 0 ∨ r = 1 := Int.emod_two_eq_zero_or_one (topCoefficient f) have hrnn : (0:ℝ) ≤ (r:ℝ) := by rcases hr01 with h | h <;> rw [h] <;> norm_num have hrle : (r:ℝ) ≤ 1 := by rcases hr01 with h | h <;> rw [h] <;> norm_num have hχabs : |parityCharacter x| = 1 := by rcases explicitApproximant_torusDistance_parityCharacter_pm x with h | h <;> rw [h] <;> norm_num have habs : |-(↑r * parityCharacter x / 2 ^ (n + 1))| = (r:ℝ) / 2 ^ (n + 1) := by rw [abs_neg, abs_div, abs_mul, abs_of_nonneg hrnn, hχabs, abs_of_pos hpos] ring have hlt : |-(↑r * parityCharacter x / 2 ^ (n + 1))| < 1 / 2 := by rw [habs, div_lt_iff₀ hpos] nlinarith
The expression rounds to 0, so its torus distance is simply its absolute value r / 2^(n + 1).
Because the absolute value is strictly less than 1/2, the value lies in (-1/2, 1/2), whose nearest integer is 0. Unfolding torusDistance and substituting round u = 0 leaves |u - 0| = |u| = r / 2^(n + 1).
uses
round_eq_zero_iff,abs_lt,Int.cast_zero,sub_zerohave hrnd : round (-(↑r * parityCharacter x / 2 ^ (n + 1))) = 0 := by rw [round_eq_zero_iff] rw [abs_lt] at hlt exact ⟨by linarith, hlt.2⟩ have hval : torusDistance (-(↑r * parityCharacter x / 2 ^ (n + 1))) = (r:ℝ) / 2 ^ (n + 1) := by unfold torusDistance rw [hrnd] simp only [Int.cast_zero, sub_zero] exact habs rw [hval]
The accepting count of f is odd if and only if r = 1.
By the parity lemma proved earlier, r = topCoefficient f % 2 equals (acceptingCount f : ℤ) % 2. An integer or natural number is odd if and only if its remainder modulo 2 is 1.
uses
explicitApproximant_torusDistance_topCoefficient_emod_two,Int.natCast_emod,Int.odd_iff,Nat.odd_iff,Nat.cast_eq_onehave hpar : r = (acceptingCount f : ℤ) % 2 := by rw [hrdef] exact explicitApproximant_torusDistance_topCoefficient_emod_two f have hcast : Odd ((acceptingCount f : ℤ)) ↔ Odd (acceptingCount f) := by have hem : ((↑(acceptingCount f) : ℤ) % 2) = (↑(acceptingCount f % 2) : ℤ) := by have hh := Int.natCast_emod (acceptingCount f) 2 simpa using hh rw [Int.odd_iff, Nat.odd_iff, hem] exact Nat.cast_eq_one have hodd : Odd (acceptingCount f) ↔ r = 1 := by rw [hpar, ← hcast, Int.odd_iff]
The torus distance equals explicitError f.
Unfolding explicitError, its value is 1 / 2^(n + 1) if acceptingCount f is odd, and 0 otherwise. If acceptingCount f is odd, r = 1, so the distance is 1 / 2^(n + 1), matching the definition. If it is not odd, r must be 0 (since r is 0 or 1), yielding distance 0, which also matches.
uses
if_pos,if_negunfold explicitError by_cases ho : Odd (acceptingCount f) · rw [if_pos ho, hodd.mp ho] push_cast ring · rw [if_neg ho] rcases hr01 with h | h · rw [h]; norm_num · exact absurd h (fun hh => ho (hodd.mpr hh))
Finset.induction_on- a property holds for all finite sets if it holds for the empty set and is preserved when adjoining an element not already present
Finset.prod_insert- the product of a function over an adjoined set insert a s is the value at a times the product over s, provided a ∉ s
Finset.univ- the finite set containing all elements of a finite type
Finset.sum_boole- the sum of an indicator condition over a finite set equals the cardinality of the subset satisfying the condition
Fintype.card_subtype- the cardinality of a subtype equals the cardinality of the corresponding finset of satisfying elements
Finset.sum_sub_distrib- the sum of differences over a finite set equals the difference of the sums
Finset.dvd_sum- if a divides each term in a finite sum, then a divides the sum
if_pos- an if-then-else expression evaluates to the 'then' branch when the condition is true
explicitApproximant_torusDistance_chi_int_pm- for any Boolean point x, the product of signs (if x i = true then 1 else -1) is either 1 or -1
if_neg- an if-then-else expression evaluates to the 'else' branch when the condition is false
Int.emod_eq_emod_iff_emod_sub_eq_zero- two integers have the same remainder modulo n if and only if their difference is divisible by n
Int.emod_eq_zero_of_dvd- if a divides b, then the remainder of b modulo a is zero
Finset.prod_eq_one- a finite product equals 1 if every factor equals 1
Finset.prod_eq_zero- in a monoid with zero, a finite product is 0 if any factor is 0
Finset.mem_univ- every element of a finite type belongs to its universal finset
round_add_intCast- rounding an integer shift satisfies round(x + m) = round(x) + m
explicitApproximant_torusDistance_fullMonomial_zero_or_one- the full monomial at any Boolean point is either 0 or 1
explicitApproximant_torusDistance_add_int- the distance to the nearest integer is invariant under integer translation: torusDistance(u + m) = torusDistance(u)
pow_le_pow_right₀- if 1 ≤ a and m ≤ n, then a^m ≤ a^n
Int.emod_two_eq_zero_or_one- the remainder of an integer modulo 2 is either 0 or 1
explicitApproximant_torusDistance_parityCharacter_pm- the real parity character at any Boolean point takes value 1 or -1
abs_neg- the absolute value of -a equals the absolute value of a
abs_div- the absolute value of a quotient is the quotient of the absolute values
abs_mul- the absolute value of a product is the product of the absolute values
abs_of_nonneg- the absolute value of a non-negative real number is the number itself
abs_of_pos- the absolute value of a strictly positive real number is the number itself
div_lt_iff₀- for c > 0, b / c < a is equivalent to b < a * c
round_eq_zero_iff- a real number rounds to 0 if and only if it lies in the half-open interval [-1/2, 1/2)
abs_lt- |a| < b is equivalent to -b < a and a < b
Int.cast_zero- the integer 0 cast to another ring equals 0
sub_zero- subtracting 0 from an element leaves it unchanged
explicitApproximant_torusDistance_topCoefficient_emod_two- the top Fourier coefficient and the accepting count of a Boolean function have the same parity
Int.natCast_emod- the integer cast of a natural remainder m % n equals the integer remainder of the casts
Int.odd_iff- an integer n is odd if and only if n % 2 = 1
Nat.odd_iff- a natural number n is odd if and only if n % 2 = 1
Nat.cast_eq_one- the cast of a natural number equals 1 if and only if the natural number is 1
Written by a model (gemini-3.8-flash) from the Lean proof and the goal states Lean reported at each step. Lean verified the proof; this explanation is a reading aid and is not itself machine-checked.
Lean source view module on GitHub
Choose a tactic — a marked line — to see the goals Lean had before and after it, as in Lean’s infoview. Step with ← → or j k.
open scoped BigOperators/-- Every product of the ±1 factors `if x i = true then 1 else -1` is `1` or `-1`. -/lemma explicitApproximant_torusDistance_chi_int_pm {n : ℕ} (x : Fin n → Bool) : (∏ i, if x i = true then (1:ℤ) else -1) = 1 ∨ (∏ i, if x i = true then (1:ℤ) else -1) = -1 := by have key : ∀ s : Finset (Fin n), (∏ i ∈ s, if x i = true then (1:ℤ) else -1) = 1 ∨ (∏ i ∈ s, if x i = true then (1:ℤ) else -1) = -1 := by intro s induction s using Finset.induction_on with | empty => simp | insert a s ha ih => rw [Finset.prod_insert ha] rcases ih with h | h <;> rcases (show (if x a = true then (1:ℤ) else -1) = 1 ∨ (if x a = true then (1:ℤ) else -1) = -1 from by split <;> simp) with h' | h' <;> rw [h', h] <;> norm_num simpa using key Finset.univ/-- The top Fourier coefficient and the accepting count of `f` have the same parity. -/lemma explicitApproximant_torusDistance_topCoefficient_emod_two {n : ℕ} (f : (Fin n → Bool) → Bool) : topCoefficient f % 2 = (acceptingCount f : ℤ) % 2 := by have hsum : (∑ x : Fin n → Bool, (if f x = true then (1:ℤ) else 0)) = (acceptingCount f : ℤ) := by rw [Finset.sum_boole] rw [acceptingCount, Fintype.card_subtype] have hdvd : (2:ℤ) ∣ topCoefficient f - (acceptingCount f : ℤ) := by have hsplit : topCoefficient f - (acceptingCount f : ℤ) = ∑ x : Fin n → Bool, ((if f x = true then ∏ i, (if x i = true then (1:ℤ) else -1) else 0) - (if f x = true then (1:ℤ) else 0)) := by rw [Finset.sum_sub_distrib, hsum] rfl rw [hsplit] refine Finset.dvd_sum ?_ intro y _ by_cases h : f y = true · rw [if_pos h, if_pos h] rcases explicitApproximant_torusDistance_chi_int_pm y with hc | hc <;> rw [hc] <;> norm_num · rw [if_neg h, if_neg h] simp exact (Int.emod_eq_emod_iff_emod_sub_eq_zero).mpr (Int.emod_eq_zero_of_dvd hdvd)/-- The parity character takes only the values `1` and `-1`. -/lemma explicitApproximant_torusDistance_parityCharacter_pm {n : ℕ} (x : Fin n → Bool) : parityCharacter x = 1 ∨ parityCharacter x = -1 := by rw [parityCharacter] have key : ∀ s : Finset (Fin n), (∏ i ∈ s, (if x i = true then (1:ℝ) else -1)) = 1 ∨ (∏ i ∈ s, (if x i = true then (1:ℝ) else -1)) = -1 := by intro s induction s using Finset.induction_on with | empty => simp | insert a s ha ih => rw [Finset.prod_insert ha] rcases ih with h | h <;> rcases (show (if x a = true then (1:ℝ) else -1) = 1 ∨ (if x a = true then (1:ℝ) else -1) = -1 from by split <;> simp) with h' | h' <;> rw [h', h] <;> norm_num simpa using key Finset.univ/-- The full monomial at a Boolean point is `0` or `1`. -/lemma explicitApproximant_torusDistance_fullMonomial_zero_or_one {n : ℕ} (x : Fin n → Bool) : fullMonomial x = 0 ∨ fullMonomial x = 1 := by by_cases h : ∀ i, x i = true · right rw [fullMonomial] refine Finset.prod_eq_one ?_ intro i _ rw [h i] simp [bit] · left push Not at h obtain ⟨i, hi⟩ := h rw [fullMonomial] refine Finset.prod_eq_zero (Finset.mem_univ i) ?_ simp [bit, hi]/-- Shifting by an integer does not change the distance to the nearest integer. -/lemma explicitApproximant_torusDistance_add_int (u : ℝ) (m : ℤ) : torusDistance (u + (m:ℝ)) = torusDistance u := by unfold torusDistance rw [round_add_intCast] congr 1 push_cast ring/-- Pointwise torus error of the explicit approximant at every Boolean point. -/theorem explicitApproximant_torusDistance {n : ℕ} (hn : 0 < n) (f : (Fin n → Bool) → Bool) (x : Fin n → Bool) : torusDistance (explicitApproximant f x - bit (f x) / 2) = explicitError f := by have hsub : explicitApproximant f x - bit (f x) / 2 = -(↑(topCoefficient f / 2) * fullMonomial x) - ↑(topCoefficient f % 2) * parityCharacter x / 2 ^ (n + 1) := by simp only [explicitApproximant, bit] ring rw [hsub] set r := topCoefficient f % 2 with hrdef set k := topCoefficient f / 2 with hkdef have hint : ∃ m : ℤ, (-(↑k * fullMonomial x) - ↑r * parityCharacter x / 2 ^ (n + 1)) = (-(↑r * parityCharacter x / 2 ^ (n + 1))) + (m : ℝ) := by rcases explicitApproximant_torusDistance_fullMonomial_zero_or_one x with h | h · refine ⟨0, ?_⟩ rw [h] push_cast ring · refine ⟨-k, ?_⟩ rw [h] push_cast ring obtain ⟨m, hm⟩ := hint rw [hm, explicitApproximant_torusDistance_add_int] have hpow : (4:ℝ) ≤ 2 ^ (n + 1) := by have h2 : 2 ≤ n + 1 := by omega calc (4:ℝ) = 2 ^ 2 := by norm_num _ ≤ 2 ^ (n + 1) := pow_le_pow_right₀ (by norm_num) h2 have hpos : (0:ℝ) < 2 ^ (n + 1) := by positivity have hr01 : r = 0 ∨ r = 1 := Int.emod_two_eq_zero_or_one (topCoefficient f) have hrnn : (0:ℝ) ≤ (r:ℝ) := by rcases hr01 with h | h <;> rw [h] <;> norm_num have hrle : (r:ℝ) ≤ 1 := by rcases hr01 with h | h <;> rw [h] <;> norm_num have hχabs : |parityCharacter x| = 1 := by rcases explicitApproximant_torusDistance_parityCharacter_pm x with h | h <;> rw [h] <;> norm_num have habs : |-(↑r * parityCharacter x / 2 ^ (n + 1))| = (r:ℝ) / 2 ^ (n + 1) := by rw [abs_neg, abs_div, abs_mul, abs_of_nonneg hrnn, hχabs, abs_of_pos hpos] ring have hlt : |-(↑r * parityCharacter x / 2 ^ (n + 1))| < 1 / 2 := by rw [habs, div_lt_iff₀ hpos] nlinarith have hrnd : round (-(↑r * parityCharacter x / 2 ^ (n + 1))) = 0 := by rw [round_eq_zero_iff] rw [abs_lt] at hlt exact ⟨by linarith, hlt.2⟩ have hval : torusDistance (-(↑r * parityCharacter x / 2 ^ (n + 1))) = (r:ℝ) / 2 ^ (n + 1) := by unfold torusDistance rw [hrnd] simp only [Int.cast_zero, sub_zero] exact habs rw [hval] have hpar : r = (acceptingCount f : ℤ) % 2 := by rw [hrdef] exact explicitApproximant_torusDistance_topCoefficient_emod_two f have hcast : Odd ((acceptingCount f : ℤ)) ↔ Odd (acceptingCount f) := by have hem : ((↑(acceptingCount f) : ℤ) % 2) = (↑(acceptingCount f % 2) : ℤ) := by have hh := Int.natCast_emod (acceptingCount f) 2 simpa using hh rw [Int.odd_iff, Nat.odd_iff, hem] exact Nat.cast_eq_one have hodd : Odd (acceptingCount f) ↔ r = 1 := by rw [hpar, ← hcast, Int.odd_iff] unfold explicitError by_cases ho : Odd (acceptingCount f) · rw [if_pos ho, hodd.mp ho] push_cast ring · rw [if_neg ho] rcases hr01 with h | h · rw [h]; norm_num · exact absurd h (fun hh => ho (hodd.mpr hh))