The truncated Chebyshev series does not generally minimize the L-inf error among all polynomials of the same degree (unless the function being approximated is a

The truncated Chebyshev series does not generally minimize the L-inf error among all polynomials of the same degree (unless the function being approximated is a polynomial of degree n+1). Constructing a minimax approximation is typically done via the Remez exchange algorithm, which is iterative, fussy, and prone to convergence failures. However, none of those matter when designing an offline approximation as is the case when you're writing math library functions.