1

I have the following sum: \begin{equation} \sum_{j=0}^{a} \sum_{k=0}^{n-2j} c_{jk}\,\, x^{\,j+k} \end{equation} Where $a=\lfloor n/2\rfloor$. I want to convert the previous sum to other like: \begin{equation} \sum_{l=0}^n a_lx^l \end{equation} but I have been struggling in the intent to writte the coefs. $a_l$ in term of the coefs $c_{ij}$. Any clue how to solve this problem will be highly appreciated.

RobPratt
  • 45,619

2 Answers2

1

Define two sets of pairs of natural numbers to serve as indices for summation: \begin{gather*} K = \{ (j, k) \colon j \geqslant 0, \ 0 \leqslant k \leqslant n - 2j \}, \\ L = \{ (j, l) \colon 0 \leqslant j \leqslant \min\{l, n - l\} \}. \end{gather*} Verify that these two functions are well defined, and are inverse to each other: \begin{gather*} \lambda \colon K \to L, \ (j, k) \mapsto (j, j + k), \\ \kappa \colon L \to K, \ (j, l) \mapsto (j, l - j), \\ \kappa \circ \lambda = 1_K, \\ \lambda \circ \kappa = 1_L. \end{gather*} Write the function to be summed as: $$ f \colon K \to \mathbb{C}, \ (j, k) \mapsto c_{jk}x^{j+k}. $$ Then: \begin{align*} \sum_{j=0}^{\left\lfloor n/2\right\rfloor}\sum_{k=0}^{n-2j} c_{jk}x^{j+k} & = \sum_{(j,k)\in K}f(j, k) \\ & = \sum_{(j,l)\in L}f(\kappa(j, l)) \\ & = \sum_{(j,l)\in L}f(j,l-j) \\ & = \sum_{(j,l)\in L}c_{j,l-j}x^l \\ & = \sum_{l=0}^n\left(\sum_{j=0}^{\min\{l,n-l\}}c_{j,l-j}\right)x^l. \end{align*}

  • I have already find "almost" the same answer in a different way. Can you tell me where is the mistake? Anyway your answer seems to be more pro. I will check it. – Federico Vega Apr 04 '20 at 19:22
  • I'll check my answer, too (but it'll take me a while, because I have to eat dinner first). I did read through it a few times before posting, but it could still easily be mistaken. I hope at least that it is systematic (not to say pedantic!) enough to make it easy to find any mistakes. – Calum Gilhooley Apr 04 '20 at 19:25
  • I left above my solution... I think your answer heavy relies on the fact that $\kappa(j,l)$ is invertible. Is it any kind of general theorem of substitution of indexes in sums? I didnt know it. I will search it. – Federico Vega Apr 04 '20 at 19:34
  • I wouldn't be surprised if Bourbaki, for example, gives an explicit statement of some such rule. I'm definitely interested in searching for it, but I can't do it today. If I can't find an explicit statement anywhere, I'll probably post a question about it, probably along with some sort of proof that I've concocted ... I have a dim memory (but my memory could be playing tricks) that there might be something in Lang's Algebra. I can have a look at that later this evening. – Calum Gilhooley Apr 04 '20 at 19:49
  • It isn't in either Lang's Algebra or Bourbaki's Algebra, Part I, not even as an exercise. Lang just mentions that there are a number of rules that it would be "tedious" to detail. (Me, I like such tedium!) – Calum Gilhooley Apr 04 '20 at 20:36
  • If I may suggest an analogy: it's like choosing a convenient "coordinate system" for evaluating a double integral; but a sum over a finite index set is much simpler than an integral. In particular, there is no need to calculate anything like a Jacobian, because all terms have weight $1.$ One need only pick a bijective "coordinate transformation", guided by the plan to evaluate the "double integral" as an "iterated integral". The much more complicated calculus problem is written up in textbooks, but the simpler discrete problem is left to whim, as if it were trivial, which it is not. Rant ends! – Calum Gilhooley Apr 04 '20 at 23:30
  • There is an intermediate level of complexity, in which some or all of the series involved may be infinite, and their terms must belong to a commutative topological group (or semigroup?). The necessary theory can be found in Bourbaki and Dieudonne, but the underlying simpler case of finite sums does not appear to be dealt with explicitly even by those authors. – Calum Gilhooley Apr 04 '20 at 23:55
  • I don't know why I thought it wasn't in Bourbaki! It was the first place I looked. It's on page 9! A PDF can be downloaded here. I'm peeved at myself for having only just noticed this (and posting a misleading comment - sorry!), because I've put a good deal of effort into answering my own question on the topic. I don't think that effort has been entirely wasted, because my work starts from a point reached in many textbooks, so doesn't require an immersion in Bourbaki's system. – Calum Gilhooley Apr 15 '20 at 16:41
0

I defined: $\bar{c}_{jk}=c_{jk}\,\,\theta\left(n-2j-k\right)$, where $\theta(x)$ is the step function ($\theta(x)=1$ if $x\geq0$ otherwise $\theta(x)=0$). So: \begin{equation} \begin{array}{c} \displaystyle \sum_{j=0}^{a} \sum_{k=0}^{n-2j} c_{jk}\,\, x^{\,(j+k)}= \sum_{j=0}^{a} \sum_{k=0}^{n} \bar{c}_{jk}\, x^{\,(j+k)} \underset{(1)}{=} \sum_{j=0}^{a} \sum_{l=0}^{a+n} \bar{c}_{j,l-j}\theta(l-j)\,x^{\,l} \underset{(2)}{=} \\ \\ \displaystyle \sum_{l=0}^{n}\sum_{j=0}^{a} \bar{c}_{j,l-j}\theta(l-j)\, x^{\,l}\underset{(3)}{=} \sum_{l=0}^{n}\sum_{j=0}^{j_0} c_{j,l-j} x^{\,l}=\sum_{l=0}^{n}a_lx^{ l}\,. \end{array} \end{equation} Where in step (1) I defined $l:=j+k$. Its clear that $0\leq l \leq a+n$. In (2) If $l\geq n$ then $\theta(n-l-j)=0$. In (3) I defined $j_0:=min\{a,n-l,l \}=min\{n-l,l \}$ and I used that $\theta(n-l-j)=0$ if $j\geq j_0$.

  • Where does the factor of $2$ in the exponent of $x$ come from? (Sorry if this is a silly question - I'm very rushed.) – Calum Gilhooley Apr 04 '20 at 19:32
  • Sorry I copy paste it from my .tex without checking that was not exacly the same summation. I have already correct it. – Federico Vega Apr 04 '20 at 20:04
  • If $n = 2$ and $l = 0,$ then $j_0 = a = 1,$ so your sum for $a_l$ contains a term $c_{1,-1},$ which is not defined. (I'll do some similar checks on my own answer now.) – Calum Gilhooley Apr 04 '20 at 21:05
  • I have been thinking what happened with my argument. The condition $k\geq0$ must be satisfied with the addition of another $\theta(l-j)$. This step function prevents the bad behaviour you noticed in the prevoius comment and it force to redifen $j_0$ as $j_0=min{n-l,l}$ :-). Even when I am happy with my demostration I still think it is very "unnatural". – Federico Vega Apr 05 '20 at 00:17