Weyl's inequality, degree
ProvedCircleMethod.weyl_inequalityThe goal of this mission. Weyl's inequality for a monomial phase of degree .
Fix . For every there is a constant such that whenever , , and ,
Proved by Weyl in 1916, this is the estimate that makes the minor arcs of the circle method negligible. Its proof is iterations of the differencing step: squaring the sum and substituting lowers the degree of the phase by one, at the cost of a square root in the final bound, which is where the exponent comes from. After steps the phase is linear and the geometric sum bound applies; the divisor bound is then used to count the differencing parameters that contribute significantly.
The estimate is only useful for in an intermediate range — roughly — and is trivially true outside it. That is exactly the point: the range where it is trivial is the major arcs, where the sum really is large and must be evaluated asymptotically rather than bounded.
Because the saving decays exponentially in , Weyl's inequality becomes weak for large degree, which is why Vinogradov's mean value method supersedes it there. For small , and as the foundation of the subject, it remains the standard tool.
Formalization Note. The phase is the pure monomial ; the general polynomial version of Vaughan's Lemma 2.4 is a strengthening and would be a welcome separate contribution. The exponents and are real powers (Real.rpow); and are natural powers. The constant is quantified after and and before , and this order must be preserved: allowing to depend on would make the statement vacuous.
import Definitions.Def_CircleMethod_genfun import Mathlib.Analysis.SpecialFunctions.Pow.Real import Mathlib.Analysis.SpecialFunctions.Integrals.Basic import Mathlib.NumberTheory.Divisors import Mathlib.Data.Fintype.Pi open Finset
namespace CircleMethod
theorem weyl_inequality {k : ℕ} (hk : 2 ≤ k) {ε : ℝ} (hε : 0 < ε) :
∃ C : ℝ, 0 < C ∧ ∀ (N q : ℕ) (a : ℤ) (α : ℝ), 0 < N → 0 < q →
Nat.Coprime a.natAbs q → |α - (a : ℝ) / (q : ℝ)| ≤ 1 / (q : ℝ) ^ 2 →
‖∑ x ∈ Finset.range N, e (α * (x : ℝ) ^ k)‖
≤ C * (N : ℝ) ^ (1 + ε) *
((1 / (q : ℝ) + 1 / (N : ℝ) + (q : ℝ) / (N : ℝ) ^ k) ^ ((2 : ℝ) ^ (1 - (k : ℝ)))) := by
sorry
end CircleMethodRead-back
What the Lean code literally says, in plain math · claude-opus-5
Read-back — CircleMethod.weyl_inequality
Notation used by the statement. The bundle defines the additive character
a complex-valued function of a real argument (the real number is coerced into before the exponential is taken); every value has modulus . The symbol applied to a complex number is its usual absolute value . Two further definitions are present in the surrounding bundle but do not occur anywhere in this statement: the generating function for a finite set , and the distance-to-nearest-integer function . Nothing about or is asserted here.
What the declaration asserts. Let be a natural number and a real number (both are supplied implicitly, i.e. inferred from the remaining data rather than written out). Assume
- , and
- .
Under exactly these two hypotheses, the claim is the existence of a real number such that and such that, for every natural number , every natural number , every integer , and every real number satisfying the four hypotheses
- ,
- ,
- (the coprimality is between the natural number , the absolute value of the integer , and ),
- (a non-strict inequality, with and computed in the real numbers),
the following bound holds:
Reading of the two sides. On the left, the index ranges over the natural numbers (the range of , which excludes itself and includes ); each such is cast to a real number, raised to the -th power, multiplied by , and fed to . Since , the term contributes . On the right, is a real-exponent power of the real number ; the bracketed quantity is formed with real division and with a natural-number power, and it is raised to the real exponent , which is itself a real-exponent power of (so , equal to when and shrinking towards as grows; it is always strictly positive and at most ).
Quantifier order and uniformity. The constant is chosen after and are fixed but before are quantified, so a single — depending only on and — must serve all admissible quadruples simultaneously. Nothing is asserted about how depends on or , no formula or algorithm for is given, and is not claimed to be unique, minimal, or effectively computable; only its existence and positivity are claimed.
Degenerate and boundary cases silently included.
- Because and are the only size constraints, all denominators , , , are nonzero, so no division-by-zero junk values arise, and the bracketed base is strictly positive (hence its real power is well defined).
- There is no hypothesis bounding from above in terms of (no condition such as or ), and no hypothesis bounding from below beyond . The case and the case are both included.
- The integer is unrestricted in sign and may be negative; it may also be , in which case hypothesis 3 forces (since ).
- When , hypothesis 3 holds for every integer and hypothesis 4 reads ; the bracketed base is then .
- Hypothesis 4 is an inequality with , so the boundary case is included. For each fixed and each , such a pair need not exist, but no emptiness arises: the hypotheses are satisfiable (e.g. , , , any ), so the universally quantified body is not vacuous.
- The conclusion is a single non-strict inequality between real numbers; no lower bound, no asymptotic statement, and no claim about the sum itself beyond this bound is made. The theorem is stated with fixed in advance, so it asserts one such inequality per pair, not a statement uniform in .
Proof status. The declaration's proof body is sorry: no proof is supplied.
Confirmed by the mission captain (proposal self-audit).