1979
DOI: 10.1090/s0025-5718-1979-0537969-8
|View full text |Cite
|
Sign up to set email alerts
|

Adaptive numerical differentiation

Abstract: Abstract.It is well known that the calculation of an accurate approximate derivative f'(x) of a nontabular function fix) on a finite-precision computer by theis a delicate task. If ft is too large, truncation errors cause poor answers, while if ft is too small, cancellation and other "rounding" errors cause poor answers. We will show that by using simple results on the nature of the asymptotic convergence of d(ft) to /', a reliable numerical method can be obtained which can yield efficiently the theoretical ma… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1
1

Citation Types

0
9
0
2

Year Published

1985
1985
2018
2018

Publication Types

Select...
5
2
1

Relationship

0
8

Authors

Journals

citations
Cited by 25 publications
(12 citation statements)
references
References 5 publications
0
9
0
2
Order By: Relevance
“…Approximations based on difference schemes are affected by truncation and rounding errors [5,28]. The truncation errors (represented in equations (8-10) with the O symbol) decrease as the stepsize tends to zero.…”
Section: Error Analysismentioning
confidence: 99%
“…Approximations based on difference schemes are affected by truncation and rounding errors [5,28]. The truncation errors (represented in equations (8-10) with the O symbol) decrease as the stepsize tends to zero.…”
Section: Error Analysismentioning
confidence: 99%
“…In practice, the second one (the classical central difference method) is often used, but the others are useful if the central difference would violate the variable bounds (for example, if the function is undefined beyond a certain bound). Copernicus includes a full set of formulas from two to eight points [43], as well as algorithms for tuning the step size h (which is critical when using finite differences) [44]. A new open source object-oriented finite-difference library (NumDiff) is also available with a variety of user-selectable methods † † † including from two to six point formulas as well as Neville's algorithm [45].…”
Section: J Gradientsmentioning
confidence: 99%
“…Отсюда следует отсутствие стати-стических данных о точности решения тестовых задач, что является недостатком сравнения итера-ционных алгоритмов. Поэтому в данной работе предлагается приводить и статистические данные о решении с максимальной точностью, а не только данные, соответствующие критериям (13). Резуль-таты численных исследований приведены в табли-цах 2 и 3 соответственно.…”
Section:  unclassified
“…Стан-дартные программы численного дифференцирова-ния нацелены на поиск конечно-разностного интервала, обеспечивающего минимальную сум-марную ошибку вычисляемого приближения. Ал-горитмы выбора интервала для центральной конечно-разностной формулы можно найти, например, в [13]. Однако для оптимизационных приложений такие программы не подходят, так как, обеспечивая избыточную точность вычисления В [14] отмечено, что замена градиентов их ко-нечно-разностными оценками лучше всего прохо-дит в квазиньютоновских методах.…”
unclassified