I want to build a Newton-Cotes formula with weights

$\int_0^1f(x)x^\alpha dx = a_0f(0) + a_1f(1) + R(f), \alpha > -1$

But, I cannot find any example, moreover I don't really know where to search or start either.

And, after I have the formula, how could I show that it makes sense and how could I find an expression for the error using an adequate derivative of f?

I need the formula and the error expression to find a formula for

$\int_0^{h} g(t)t^\alpha dt $, for a small h > 0


The idea behind Newton-Cotes rule is basically interpolation of the integrand. In case of weighted formula, we can interpolate only $f(x)$.

The interpolation nodes for Newton-Cotes are equally spaced and contain endpoints. Writing $f(x)$ interpolant in terms of Lagrange basis interpolants we have $$ f(x) \sim P(x) \sum_{i = 1}^{K} f(x_i) \ell_i(x), $$ where $x_i = a + \frac{i-1}{K-1}(b-a)$ is a set of evenly spaced $K$ points spanning the $[a,b]$ interval and $\ell_i(x)$ is defined as $$ \ell_i(x) = \prod_{j \neq i} \frac{x - x_j}{x_i - x_j} $$ Substituting $P(x)$ for $f(x)$ in the integral $$ \int_a^b f(x) x^\alpha dx \approx \int_a^b P(x) x^\alpha dx = \sum_{i=1}^K w_i f(x_i) $$ with $w_i$ being $$ w_i \equiv \int_a^b \ell_i(x) x^\alpha dx. $$ The interpolation error could be roughly estimated as $$ \left|P(x) - f(x)\right| \leq \frac{1}{K!}\max\limits_{z \in [a,b]} \left|f^{(K)}(z)\right| \left|\prod_{i=1}^K (x-x_i)\right| $$ therefore quadrature error is bounded by $$ \varepsilon = \left|\int_a^b (f(x) - P(x))x^\alpha dx\right| \leq \int_a^b \left|f(x) - P(x)\right| x^\alpha dx \leq \frac{1}{K!}\max\limits_{z \in [a,b]} \left|f^{(K)}(z)\right| \int_a^b \left|\prod_{i=1}^K (x-x_i)\right| x^\alpha dx $$

For your case $K = 2$ and $a = 0, b = 1$. Lagrange basis polynomials are $\ell_1 = 1 - x$ and $\ell_2 = x$, so $$ w_1 = \int_0^1 (1-x)x^\alpha dx = \frac{1}{\alpha ^2+3 \alpha +2}\\ w_2 = \int_0^1 x \cdot x^\alpha dx = \frac{1}{\alpha + 2}\\ \varepsilon \leq \frac{1}{2}\max\limits_{z \in [a,b]} \left|f''(z)\right| \int_0^1 |(x-0)(x-1)| x^\alpha dx = \\ = \frac{1}{2}\max\limits_{z \in [a,b]} \left|f''(z)\right| \int_0^1 x(1-x) x^\alpha dx = \frac{1}{12+10\alpha +2\alpha^2}\max\limits_{z \in [a,b]} \left|f''(z)\right| $$ There are more precise ways to esimate quadrature errors, ask if you need them.

Simple test case. Let $\alpha = \frac{1}{2}$, $f(x) = \cos x$. $$ w_1 = \frac{4}{15}, \qquad w_2 = \frac{6}{15}\\ \int_0^1 \cos x\sqrt{x} dx = 0.5312026830...\\ \sum_{i=1}^2 w_i f(x_i) = w_1 \cos 0 + w_2 \cos 1 = 0.4827875890... $$ The actual error is $\varepsilon = 0.048415$ and the error estimate is $$ \varepsilon \leq \frac{2}{35} \max\limits_{z \in [0, 1]} \left|-\cos z\right| = \frac{2}{35} \approx 0.0571429. $$


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.