So the estimation error is introduced at the step where a function is approximated with another function, which is usually chosen as either a polynomial or a polynomial spline (composed of straight line segments for the simplest trapezoidal integration), not at the actual integration.
Fortunately, for well-behaved functions, when they are approximated by a suitable simpler function, the errors that this approximation introduces in the values of function integrals are smaller than the errors in the interpolated values of the function, which are in turn smaller than the errors in the values estimated at some point for the derivative of the function (using the same approximating simpler function).
The key property of quadrature formulas (i.e. numerical integration formulas) is the degree of exactness, which just says up to which degree we can integrate polynomials exactly. The (convergence of the) error of the quadrature depends on this exactness degree.
If you approximate the integral using a sum of n+1 weights and function evaluations, then any quadrature that has exactness degree n or better is in fact an interpolatory quadrature, that is, it is equivalent to interpolating your function on the n+1 nodes and integrating the polynomial. You can check this by (exactly) integrating the Lagrange basis polynomials, through which you can express the interpolation polynomial.
> That is to say, with n nodes, gaussian integration will approximate your function's integral with a higher order polynomial than a basic technique would - resulting in more accuracy.
This is not really the case, Gaussian integration is still just interpolation on n nodes, but the way of choosing the nodes increases the integration's exactness degree to 2n-1. It's actually more interesting that Gaussian integration does not require any more work in terms of interpolation, but we just choose our nodes better. (Actually, computing the nodes is sometimes more work, but we can do that once and use them forever.)
---EDIT---
I'm about 98% sure this blog has a browser hijack embedded in it targeted at windows+MSEDGE browsers that attempted to launch a malicious powershell script to covertly screen record the target machine
The code for my blog is here : https://github.com/RohanGautam/rohangautam.github.io
If you could point to anything specific to support that claim, would be nice.
I'll try to get more details.
I should note, I do not believe the site is malicious, but i am worried about 3rd party compromise of the site without the owner's knowledge
What is also worth pointing out and which was somewhat glanced over is the close connection between the weight function and the polynomials. For different weight functions you get different classes of orthogonal polynomials. Orthogonal has to be understood in relation to the scalar product given by integrating with respect to the weight function as well.
Interestingly Gauss-Hermite integrates on the entire real line, so from -infinity to infinity. So the choice of weight function also influences the choice of integration domain.
Like, is it possible to infer that Chebyshev polynomials would be useful in approximation theory using only the fact that they're orthogonal wrt the Wigner semicircle (U_n) or arcsine (T_n) distribution?
The weight function shows the Chebyshev polynomials' relation to the Fourier series . But they are not what you would usually think of as a good candidate for L2 approximation on the interval. Normally you'd use Legendre polynomials, since they have w = 1, but they are a much less convenient basis than Chebyshev for numerics.
If you are familiar with the Fourier series, the same principle can be applied to approximating with polynomials.
In both cases the crucial point is that you can form an orthogonal subspace, onto which you can project the function to be approximated.
For polynomials it is this: https://en.m.wikipedia.org/wiki/Polynomial_chaos
That is, the integral from - to + infinity of e^(-x^2) dx = sqrt(pi).
I remember being given this as an exercise and just being totally shocked by how beautiful it was as a result (when I eventually managed to work out how to evaluate it).
It's the gateway drug to Laplace's method (Laplace approximation), mean field theory, perturbation theory, ... QFT.
It would be better shown as a table with 3 numbers. Or, maybe two columns, one for integral value and one for error, as you suggest.
(For some reason, plt.bar was used instead of plt.plot, so the y axis would start at 0 by default, making all results look the same. But when the log scale is applied, the lower y limit becomes the data’s minimum. So, because the dynamic range is so low, the end result is visually identical to having just set y limits using the original linear scale).
Anyhow for anyone interested the values for those 3 points are 2.0000 (exact), 1.9671 (trapezoid), and 1.9998 (gaussian). The relatives errors are 1.6% vs. 0.01%.
At first glance I see we're talking about numerical integration so I assuming the red part is this method that is being discussed and that it's much better than the other two. Then I look at the axis, which the caption notes is log scale, and see that it goes from 1.995 to 2. Uh-oh, is someone trying to inflate performance by cutting off the origin? Big no-no, but then wait a minute, this is ground truth compared to two approximations. So actually the one on the right is better. But the middle one is still accurate to within 0.25%. And why is it log scale?
Anyway point is, there's lots of room for improvement!
In particular it's probably not worth putting ground truth as a separate bar, just plot the error of the two methods, then you don't need to cut off the original. And ditch the log scale.
