The page is a *working tool* for finding optimal evaluation chains for a specific polynomial of degree 9. You input a polynomial, pick a target field (rationals, reals, complex, or finite fields), and the system generates evaluation methods that minimize multiplications, additions, or critical-path depth. The main claim: their "This paper" method achieves 5 multiplications and 4-depth with *exact* rational arithmetic—no numerical error—by doing rational preprocessing on the polynomial coefficients ahead of time. For comparison, Horner uses 8 multiplications sequentially (depth 8), Estrin parallelizes to depth 4 but needs 11 multiplications, and Pan's 1978 scheme also hits 5 multiplications but only works numerically. The output is a concrete evaluation chain in mathematical notation and C code you can copy and run. The trade-off is upfront work to find the best factorization; the payoff is fewer operations or shallower circuits for runtime evaluation. The source doesn't explain *why* rational preprocessing works or how the tool finds these chains algorithmically—that's apparently in their arxiv paper (2609.06022, dated 2026—likely a timestamp artifact).
reply