Inspired by this quiz about discrete derivatives.
When I found this quiz, I remembered a thought I had had when I had heard about the formula ∑k=1nk=2n(n+1).
If you expand the fraction, you get 21n2+21n, which almost looks like the antiderivative of k except for the 21n term. When I did this for other sums, I got the same result:
∑k=1nk2=31n3+21n2+61n
∑k=1nk3=41n4+21n3+41n2
∑k=1nk4=51n5+21n4+31n3−301n
...
Because of the definition of these sums, their discrete derivatives are always na (a is some natural number), but even for discrete derivatives of simple polynomials the fundametal theorem of calculus still almost holds; at least for the highest order term.
So now I'm wondering if
- this also holds for functions that aren't polynomials, for example exponential functions or trigonometric functions
- the discrete antiderivative of the discrete antiderivative is also similar to the normal second antiderivative
- there are some polynomials for which antiderivative and discrete antiderivative are exactly the same
EDIT: I've found the thing I'm calling "discrete antiderivative" as finite difference on Wikipedia and the article points out Faulhaber's formulas and the difference between odd and even degree polynomials. However, I haven't found what I'm searching for – a comparison/link to standard calculus integrals.
#Combinatorics
Easy Math Editor
This discussion board is a place to discuss our Daily Challenges and the math and science related to those challenges. Explanations are more than just a solution — they should explain the steps and thinking strategies that you used to obtain the solution. Comments should further the discussion of math and science.
When posting on Brilliant:
*italics*
or_italics_
**bold**
or__bold__
paragraph 1
paragraph 2
[example link](https://brilliant.org)
> This is a quote
\(
...\)
or\[
...\]
to ensure proper formatting.2 \times 3
2^{34}
a_{i-1}
\frac{2}{3}
\sqrt{2}
\sum_{i=1}^3
\sin \theta
\boxed{123}
Comments
The formulas you have provided are called Faulhaber's formulas and are linked to Bernoulli numbers. There is a result by John Conway recently (2016, I think), where he is able to compute "derivatives" and "integrals" of these Faulhaber formulas, which distinguish between odd degrees and even degrees of the polynomials. Note that here, the notions of "derivative" and "integral" are merely just linear functions on polynomials in n, which sends one polynomial to another with the linearity properties satisfied; no notion of calculus is required here. Because of this, we need to be precise as to what a "derivative" and an "integral" is here, as well as what a "discrete derivative" and a "discrete antiderivative" here. None of your questions can be answered without first establishing this point; you cannot simply input the framework of calculus into something that is only applied in a discrete way.
P.S. As for "exponential" functions and "trigonometric" functions, I don't see the relevance of Faulhaber's results to them as they are not properly defined currently.
Log in to reply
I would say,
Then all those four functions applied to polynomials are also polynomials and satisfy linearity (because of the commutative property for 2. and the linearity of sums for 3.)
So with these definitions, I hope I have stated my questions unambiguously, but I also think that they don't really make sense for other functions like exponential or trigonometric functions. But Conway's result sounds interesting, I think I'll take a look at it.
Log in to reply
What power rule from calculus? It's simplest to just define the Derivative as a linear map, where for integers n≥1,
D(xn)≡nxn−1
and the Integral as the inverse mapping of the Derivative, i.e. for integers n≥0,
I(xn)≡D−1(xn)=n+1xn+1.
This was how Conway presented his definitions; with this, you don't need to be sloppy with them and avoids having to define a "discrete" version of them. As I said, no notion of calculus is ever involved; we are merely defining the Derivative and Integral as linear maps/functions which map one polynomial to another.
My main point was to ask you why you care to link these to the "standard" framework of calculus, when we have a nice stand-alone framework to study Faulhaber polynomials which only involve algebra & combinatorics.
P.S. I urge you again to look up Bernoulli numbers; I think this should answer your question regarding the coefficient of the largest-degree term.
Log in to reply
And then I also thought, if we have one mapping I(xn)=D−1(xn)=n+1xn+1 and then another mapping F(xn)=(Faulhaber′sformula), of course it will be interesting if there exist polynomials for which these two mappings give the same result.
Log in to reply
As to your final sentence, I vehemently disagree; there is no polynomial that is invariant under the Derivative or Integral linear map (it is easy to show this). You will have to venture into the realm of fiction to find such a case, by relying on "infinite processes" which would go against everything in discrete mathematics.
EDIT: To save you the trouble, note that what you would call "exponential" or "trigonometric" functions are really just "infinite sums".
Log in to reply
p(x) for which I(p(x))=F(p(x)) is true?
Could you maybe describe a little proof why there exists no polynomialI don't directly see an intuitive reason why this is true, but I also haven't found an example.
Log in to reply
I think you are asking the wrong questions. Did you read up on Faulhaber polynomials? They simply give the sum of the kth powers of the first n numbers... so it does not make sense to talk about the Faulhaber polynomial of a polynomial.
I firmly believe you are trying to muddy the waters between the standard framework of calculus and what one would call "discrete calculus", which relies more on difference equations than limiting processes. The two areas, while not mutually exclusive, are highly divergent from one another and thus it's important to separate both frameworks.
Log in to reply
Of course
and
Now I thougut, every polynomial (like p(n)) is a linear combination of all powers of n. So the Faulhaber polynomial could be the same linear combination of the individual Faulhaber polynomials.
This makes sense because F(na) is just a partial sum of na and, since sums are linear, we can split one sum of a polynomial into many sums, one for each term of the polynomial.
In my example, I would argue
Log in to reply
Log in to reply
2F(n2−2n) and I think this is equal to t he right side with doubled coefficients. Or isn't that the definition of linear?
I think it is. If you double the left side, you haveLog in to reply
Log in to reply
Wikipeda says