[HN Gopher] Notes on Taylor and Maclaurin Series
       ___________________________________________________________________
        
       Notes on Taylor and Maclaurin Series
        
       Author : ibobev
       Score  : 65 points
       Date   : 2024-07-24 13:30 UTC (5 days ago)
        
 (HTM) web link (eli.thegreenplace.net)
 (TXT) w3m dump (eli.thegreenplace.net)
        
       | zamadatix wrote:
       | All interests I had in Taylor/Macluarin series were promptly beat
       | out of me when having to manually calculate hundreds of them in
       | Calculus courses :).
       | 
       | The Remez algorithm linked at the end still inspires curiosity
       | though. Any other "numerically more useful" approximation
       | algorithms folks want to highlight? The Pade approximant looks
       | like another interesting candidate to read about.
        
         | raphlinus wrote:
         | Yes! The workhorse numerical technique is Chebyshev
         | approximation. Remez exchange usually starts with it, for fine-
         | tuning with respect to a "maximum error" norm, but it also
         | works quite well by itself, and can be computed efficiently
         | (even with a high degree) even from a function you can only
         | evaluate numerically.
         | 
         | A really good place to read up on it is the documentation for
         | Chebfun.
         | 
         | https://www.chebfun.org/docs/guide/guide04.html
         | 
         | Also: be on the lookout for a blog post on using Chebyshev
         | polynomials to efficiently compute error metrics for curves.
        
           | bee_rider wrote:
           | Chebyshev's polynomials seem to have eclipsed his semi-
           | iterative method for solving linear systems, which is too bad
           | IMO.
        
         | bee_rider wrote:
         | I had the opposite reaction.
         | 
         | Math department: "oh, look, we are very clever, see these
         | pretty plots the differential equations generate and all the
         | beautiful closed form solution we can make"
         | 
         | Me: "Hmm, very nice, I am not smart enough for this."
         | 
         | Engineering department: "Complicated equation bad, smash with
         | Taylor series, little part go bye-bye, big part is smooth."
         | 
         | Me: "Yes this equation will match my brain nicely."
        
         | abdullahkhalids wrote:
         | Weird. When I was doing boxing, having to run sprints every
         | training session, doing dozens of pushups, situps and what not
         | did not beat the interest I had in boxing :).
        
           | borroka wrote:
           | Not at all weird, because you were also boxing. But if your
           | boxing sessions were only to be sprints, pushups, situps etc,
           | and no boxing, you would surely lose interest in doing boxing
           | at that place. Likewise, if your jiu jitsu training were only
           | drills and no sparring, there would be no people doing jiu
           | jitsu after a while.
        
             | abdullahkhalids wrote:
             | My boxing training sessions were about 50-50 exercises and
             | the padding/sparring session (in which there were big
             | breaks).
             | 
             | In my undergrad, I was made to take 6 math courses, and
             | about 18 physics courses which used that math [1]. Plus
             | additional Engineering/CS electives. So over a 25-75 ratio
             | of exercise to "real" stuff.
             | 
             | Much better ratio than boxing.
             | 
             | [1] If you decide that the math courses have nothing
             | real/interesting in them. Which is not true.
        
       | brianm wrote:
       | A coworker, once, had a cool idea to use Taylor Series to encode
       | histograms for metrics collection. Basically a digest method akin
       | to sketches or t-digests. We wound up using t-digests as Stripe
       | (iirc) had a good OSS implementation at the time, but using
       | Taylor Series has been lingering in my mind ever since.
        
         | curt15 wrote:
         | Are you referring to generating functions?
        
       | ziofill wrote:
       | For some functions you can turn the Taylor series into a
       | recurrence relation, which makes it blazing fast to calculate.
       | 
       | For instance for the family of functions f(x0,x1,...) =
       | exp(poly(x0,x1,...)), where poly is a multivariate polynomial of
       | order m, you can compute the Taylor coefficients with a
       | recurrence relation of order m (that needs to look back m steps).
       | This shows up in quantum optics, for example.
        
       | skzv wrote:
       | One of the most useful things I learned in mathematics.
       | 
       | Gauss-Newton algorithm, GNSS (GPS) solution solving, financial
       | calculus (Ito calculus), and more.
        
       | btilly wrote:
       | In complex analysis, all differentiable (in a region) functions
       | are infinitely differentiable, and are the same as their Taylor
       | series.
       | 
       | Real analysis is a zoo of weird exceptions. Including
       | 1/e^(-1/x^2) away from 0, 0 at 0. Its Maclaurin series is just 0,
       | which is clearly not the function we wrote down.
       | 
       | I can't explain why real analysis fit my brain and complex
       | analysis doesn't. But to me complex analysis looks like, "We draw
       | a path, then calculate this contour integral, and magic happens."
        
         | curt15 wrote:
         | Instead of thinking of real analysis as a zoo of weird
         | exceptions, it's probably more accurate to think of complex
         | analyic functions as the exceptions. For example, when viewed
         | as a two dimensional mapping from the plane to itself, complex-
         | analytic functions are conformal (angle-preserving) whereas
         | most differentiable mappings from the plane to itself are not.
        
           | dhosek wrote:
           | On the other hand, complex analysis explains things that
           | aren't obvious in real analysis, like why the radius of
           | convergence for a continuous function on the real line might
           | be 1 (turns out there are singularities off the real line on
           | the complex plane).
        
         | kaashif wrote:
         | Complex analysis is far too nice. There are so many theorems
         | where my first thought was "there's no way that's true".
         | 
         | Picard's great theorem is totally insane. As is even its little
         | brother, and even Liouville's theorem.
         | 
         | The proofs aren't even that long. They just feel totally false.
        
       | antognini wrote:
       | One of the things that has always seemed rather magical to me is
       | the Taylor series of the exponential function for very large
       | negative values. We know that a number like e^-100 is a number
       | extremely close to zero. Yet when you write out the Taylor series
       | you see an alternating some of increasingly large, seemingly
       | arbitrary numbers (at least for ~100 terms):
       | 
       | e^-100 = 1 - 100 + 5000 - 166,666.6 + 4,166,666.6 - ...
       | 
       | If you were just given this sum and knew nothing about Taylor
       | series or the exponential function you'd assume that its value
       | was some extremely large number. Yet everything manages to cancel
       | out just right so that the resulting sum is almost exactly zero,
       | but not quite.
       | 
       | I can't help but wonder if there's some parallel to parts of
       | quantum field theory. If you expand out the interactions between
       | two particles you also end up with a series that is apparently
       | divergent. Yet we know experimentally that the first few terms
       | work quite well as an approximation. It feels a bit like you're
       | looking at the series of e^-100 without knowing about the
       | exponential function.
        
       ___________________________________________________________________
       (page generated 2024-07-29 23:00 UTC)