Total control of the quadratic Siegel convolution discrepancy #
The floor sum is split at m. On d < m, the floor error is at most one
and |Re χ(d)| ≤ 1, so the complete short contribution costs at most m.
On d ≥ m, the existing Pólya--Vinogradov/Abel estimate controls the
floor-weighted tail. The final theorem inserts this honest total estimate
back into the square-lower-bound/Pólya--Vinogradov bridge.
This closes the discrepancy-control layer. It does not by itself prove a
large-conductor Siegel lower bound: optimizing this one-fold square estimate
leaves an error on the scale forced by sqrt q (1 + log q). A genuine
higher-convolution or power-amplification producer is still required before
one can derive L(1,χ) ≫_η q⁻η.
Total discrepancy bound: short floor error plus the oscillatory long tail.
The square lower bound with the total discrepancy estimate substituted. This is the strongest unconditional output of the present one-fold route.