Skip to the argument
Proof Atlas

Where the windows come from

The source constructs intervals, compares a weighted prime sum with geometric data, and absorbs the errors into a power just above one half.

Fix the target before choosing the data

Choose ½ < θ < 1. The arithmetic argument uses a small positive ε, a family of auxiliary objects and two loss parameters eD and eL. For the witnesses supplied on the path from Main, these parameters are 2 and 0. The resulting exponent is ½ + 9ε. We trace how each part enters below.

This is a reconstruction of the submitted derivation. Its initial inputs include results on theta lifting, Galois representations, local matching, volumes and character sums. The links identify where the source supplies or uses those results. Independent proofs of those inputs would be needed for a self-contained arithmetic argument.

What survives a theta lift?

Alpöge’s note singles out iterating theta lifts about √x times. The code gives a more precise statement. For the scale x and sufficiently small ε, it chooses an even integer m satisfying

x½ − ε/16 ≤ m ≤ 2x½ − ε/16.

The rank condition bounds m between √(xδ₁) and twice that quantity. The construction sets xδ₁ = x1−ε/8. This explains the exponent and the role of the factor two in choosing an even integer. Author’s description of the iteration

At step k ≤ m, the induction preserves two pieces of information. The new vector has the prescribed parameter for stage k, and its pairing against any test vector equals the original pairing after applying the composed theta operator. At the successor step the proof calls the single-step lift with a nonvanishing hypothesis and composes the two pairing identities. This explains what the iteration transports. The family ties the selected points to the same rank m. A later divisibility estimate uses a factor proportional to m at each selected scale. The pairing identity, rank constraint and divisibility estimate play different roles in the final error bound.

Where the two arithmetic inputs meet

  1. Supply intervals. Theta-lift existence, geometric realization, congruences and descent supply a local-to-global compatibility result for Galois representations. The source combines this result with base change and other automorphic inputs to obtain its window condition, VWin.
  2. Control the objects on those intervals. Agreement of theta integrals and local matching enter the construction of evaluation data. Covolume, divisibility, rank and other estimates then link the geometry to data used by the character sum. In this part of the proof the source fixes eD = 2 and eL = 0.

Obtain an entire window

For a primitive character, the window condition and an auxiliary existence theorem for prime pairs give the following order of choices. After ε and the fixed character data are chosen, for every threshold there is an interval [Y, Y3/2] such that for each x in the interval there are suitable arithmetic data. The auxiliary primes and fields may vary with x. The construction supplies compatible hosts, meaning auxiliary objects on which the character and geometric estimates are both valid.

A separate selection step makes the bounds hold on one family for the same ε. It takes the smaller of the admissible ε thresholds, selects the auxiliary conditions, then uses existence to find a family satisfying them together. This order lets the final inequality apply at every x in the chosen window.

Account for the exponent

StepWhat controls the error
Bound each classSubtract the predicted main contribution. The local and global estimates bound that gap by an explicit error expression times xCε, with C = eD + eL + 3 = 5.
Sum the classesApply the character and sparsity estimates with the common m. The main error is at most x½+(C+3)ε, plus a remaining character term. Here C + 3 = 8.
Absorb the remainderThe remainder is bounded by a fixed level-dependent factor times x½+ε. Increase the lower threshold until 1 + 4o ≤ xε, where o is the level’s finite group order. This gives the ninth ε.

The same m used at the selected points appears in the core estimate, which requires m ≤ x½+ε/2 and xm ≤ k for its scale parameter k. The final window inequality takes the largest of the needed lower thresholds. For example, to reach θ = 0.60, choose ε no greater than 1/90 and small enough for the family construction. The same ε then applies across the whole interval.

Recover the zeta bound

Use the primitive trivial character of modulus one. Its cyclic character group has just one element. The level’s sum over nontrivial characters is empty, so the weighted level error reduces to G. Choose ε below the construction’s thresholds and with 9ε ≤ θ−½. Given any desired starting point, choose the window beyond it, beyond 1, and beyond the error thresholds. At every point in that window,

G(x) ≤ x½+9ε ≤ xθ.

This supplies precisely the hypothesis used by the analytic criterion, subject to the arithmetic results invoked above. Specializing the final level removes the nontrivial characters from the final sum. Auxiliary characters still enter the construction of the hosts and the comparison estimates.

What still needs a shorter proof?

The source explains the length of the chain and what each lift preserves. The factor 9ε comes from aggregating later estimates. A shorter arithmetic proof must show how to bound the class gaps and the remainder for the zeta level with fewer inputs. The pairing identity alone gives neither estimate.

Lean source and verification