Given the sequence $\{ f(i) \}_{0 \leq i \leq n}$ and the array $\{ g_{n}(i,j) \}_{0 \leq i, j \leq n}$ such that $f(i) = \sum_{j = 0}^{n} h_{j} \, g_{n}(i,j)$ for some integers $h_{0}, \dots, h_{n}$, is there a way to write $h_{j}$ as a sum over $f(i)$ and $g_{n}(i,j)$ using the calculus of finite differences?


Update: The first post below answers the question of writing $h_{j}$ as a function of $f$ and $g$, but I'd like to avoid matrix inversion if at all possible, hence calculus of finite differences.

  • $\begingroup$ Keeping all the parameters $g_n(i,j)$ fixed (and the matrix $g$ defined in my post invertible), any formula for $h_0$ as a function of the parameters $f(i)$ valid for every $(f(i))_i$ must be the one I gave in my answer. So, you could explain a little more how calculus of finite differences, or any other sweet technology, could enter the picture at all. $\endgroup$ – Did Mar 7 '11 at 6:39

Using the vector notations $f=(f(i))_i$ and $h=(h_i)_i$ and the matrix notation $g=(g_n(i,j))_{i,j}$, you impose that $f=g\cdot h$ and you ask for an expression of $h$ as a function of $f$ and $g$. Let $D=\det g$. If $g$ is not invertible, that is, if $D=0$, either this is impossible or the solution is not unique. If $g$ is invertible, that is, if $D\ne 0$, the unique solution is $h=g^{-1}\cdot f$ and you know that $D\cdot g^{-1}$ is the comatrix $c$ of $g$.

Thus, in the invertible case $h_i=\displaystyle\frac1D\sum_{j=0}^nc(i,j)f(j)$ for every $0\le i\le n$.

  • $\begingroup$ Excellent post! $\endgroup$ – user02138 Mar 6 '11 at 21:23
  • $\begingroup$ @user02138 Then you might want to accept it. $\endgroup$ – Did Mar 11 '11 at 22:29
  • $\begingroup$ Sure. ;) $\endgroup$ – user02138 Mar 11 '11 at 23:39

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.