Knuth–Eve algorithm
In computer science, the Knuth–Eve algorithm is an algorithm for polynomial evaluation. It preprocesses the coefficients of the polynomial to reduce the number of multiplications required at runtime.
Ideas used in the algorithm were originally proposed by Donald Knuth in 1962. His procedure opportunistically exploits structure in the polynomial being evaluated.[1] In 1964, James Eve determined for which polynomials this structure exists, and gave a simple method of "preconditioning" polynomials (explained below) to endow them with that structure.[2][note 1]
Algorithm
[edit]Preliminaries
[edit]Consider an arbitrary polynomial of degree . Assume that . Define such that: if is odd then , and if is even then .[2]
Unless otherwise stated, all variables in this article represent either real numbers or univariate polynomials with real coefficients.[1][2] All operations in this article are done over .[2]
Again, the goal is to create an algorithm that returns given any . The algorithm is allowed to depend on the polynomial itself, since its coefficients are known in advance.[1]
Overview
[edit]Key idea
[edit]Using polynomial long division, we can write
where is the divisor. Picking a value for fixes both the quotient and the coefficients in the remainder and . The key idea is to cleverly choose such that , so that
[4] This way, no operations are needed to compute the remainder polynomial, since it's just a constant. We apply this procedure recursively to , expressing
After recursive calls, the quotient is either a linear or a quadratic polynomial. In this base case, the polynomial can be evaluated with (say) Horner's method.[1][4][5]
"Preconditioning"
[edit]For arbitrary , it may not be possible to force at every step of the recursion.[1] Consider the polynomials and with coefficients taken from the even and odd terms of respectively, so that
If every root of is real, then it is possible to write in the form given above. Each is a different root of , counting multiple roots as distinct.[4] Furthermore, if at least roots of lie in one half of the complex plane, then every root of is real.[2]
Ultimately, it may be necessary to "precondition" by shifting it — by setting for some — to endow it with the structure that most of its roots lie in one half of the complex plane. At runtime, this shift has to be "undone" by first setting .[2]
Preprocessing step
[edit]The following algorithm is run once for a given polynomial .[1][4] At this point, the values of that will be evaluated on are not known.[1]
- Let be the complex roots of , sorted in descending order by real part
- Choose any
- Set
- Let and be the polynomials such that
- Let be the roots of . All of its roots will be real.
- Initialize
- For :
- Divide by to get quotient and remainder . The remainder will be a constant polynomial — a number.
- Set
- Output: The derived values , , and ; as well as the base-case polynomial
Better choice of t
[edit]While any can work, it is possible to remove one addition during evaluation if is also chosen such that two roots of are symmetric about the origin. In that case, can be chosen such that the shifted polynomial has a factor of , so . It is always possible to find such a .[2]
One possible algorithm for choosing is:[citation needed]
- If :
- If :
- Else:
- Else:
Evaluation step
[edit]The following algorithm evaluates at some, now known, point .[1][2][4][5]
Assuming is chosen optimally, . So, the final iteration of the loop can instead run
saving an addition.[2]
Analysis
[edit]In total, evaluation using the Knuth–Eve algorithm for a polynomial of degree requires additions and multiplications, assuming is chosen optimally.[2]
No algorithm to evaluate a given polynomial of degree can use fewer than additions or fewer than multiplications during evaluation. This result assumes only addition and multiplication are allowed during both preprocessing and evaluation.[6][better source needed]
The Knuth–Eve algorithm is not well-conditioned.[7]
Footnotes
[edit]References
[edit]- ^ a b c d e f g h Knuth, Donald (December 1962). "Evaluation of polynomials by computer". Communications of the ACM. 5 (12): 595–599. doi:10.1145/355580.369074. Retrieved 25 July 2025.
- ^ a b c d e f g h i j Eve, James (December 1964). "The evaluation of polynomials". Numerische Mathematik. 6 (1): 17–21. doi:10.1007/BF01386049. Retrieved 25 July 2025.
- ^ "James Eve (Newcastle University)". Author DO Series. ACM Digital Library. doi:10.1145/contrib-81100250587/abs (inactive 31 July 2025). Retrieved 2025-07-30.
{{cite web}}
: CS1 maint: DOI inactive as of July 2025 (link) - ^ a b c d e Overill, Richard (12 June 1997). "Data parallel evaluation of univariate polynomials by the Knuth-Eve algorithm". Parallel Computing. 23 (13): 2115–2127. doi:10.1016/S0167-8191(97)00096-3. Retrieved 25 July 2025.
- ^ a b Muller, Jean-Michel (17 November 2016). Elementary functions: Algorithms and implementation. Boston, MA: Birkhäuser Boston. pp. 82–84. doi:10.1007/978-1-4899-7983-4_5. ISBN 978-1-4899-7983-4. Retrieved 25 July 2025.
- ^ Erickson, Jeff (10 March 2003). "Evaluating polynomials" (PDF). CS 497: Concrete Models of Computation. University of Illinois Urbana-Champaign. Retrieved 25 July 2025.
- ^ Mesztenyi, Charles (January 1967). "Stable evaluation of polynomials". Journal of Research of the National Bureau of Standards - B. Mathematics and Mathematical Physics. 71B (1): 11–17. doi:10.6028/jres.071B.003. Retrieved 25 July 2025.