MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I've been musing about the following question for a while now. Given an integral operator $G$ defined by $$ (Gf)(x) = \int_0^1 G(x,u) f(u)\,du $$ Is it possible to decompose this into two separate "one-sided" integral operators $L$ and $R$ such that $$ (Gf)(x) = \int_x^1 L(x,u) \int_0^x R(u,v) f(v)\,dv\,du $$ If so, under what assumptions on $G$? And are there expressions for the left and right operators $L$ and $R$ in terms of $G$?

share|cite|improve this question
So you want a continuous analog of the decomposition of matrices into lower and upper triangular matrices. I'm not a specialist, but it seems to me that for reasonably nice kernels (e.g. Hilbert-Schmidt), the natural approach should be: pass to the limit in the decomposition for finite dimensional approximations, and show that the sequence of factors admits a converging subsequence. – Pietro Majer Feb 16 '12 at 9:58
@Pietro Majer: Matrixes can not be decomposed in a product of upper triangular and lower triangular matrices (look up Bruhat decomposition). You have to allow permutation matrices. But nice interpretation of the question. – Marc Palm Feb 16 '12 at 13:53
@Uday: The upper limit is $x$. @Pietro: This was indeed inspired by the LU decomposition. I'm wondering how to construct the analogous functional form of the decomposition. – Victor Liu Feb 16 '12 at 19:18
@MarcPalm: of course, I'not saying that. The OP ask under what assumptions on G it is possible; I'm saying that the infinite dimensional problem includes the analogous problem for matrices, etc. – Pietro Majer Sep 27 '13 at 18:17
The upper limit $x$ screws the problem up because then $Gf(x)$ depends only on $f(t)$ with $t\le x$, which is possible only if the operator is triangular from the beginning, i.e., when there is nothing to do really. You'd better put $u$ there as Uday suggested. – fedja Oct 28 '13 at 1:52

imho, if you can rewrite the equation in a suitable basis, you obtain an infinite matrix and may try the QR decomposition. Try wavelet basis or a parent of this family of functions like curvelets, ridglets... regarding properties of the operators $G$. If $G$ is a pseudo or Fourier integral operator, these bases are known to give almost diagonal representations.

share|cite|improve this answer

To mimic multiplication, the middle integral you wrote down should be from $0$ to $u$, not from $0$ to $x$. The way you have written it, $f(v)$ does not affect $Gf(x)$ unless $v \leq x$.

Assuming $G, L,$ and $R$ are sufficiently nice smooth functions, we should have some kind of uniqueness result where an equality of integral operators is an equality of integrals, and so we can write

$G(x,v) = \int_{\max (x,v)}^1 L(x,u) R(u,v) du$

If $L$ and $R$ are honest bounded functions, this gives $G(x,v)=0$ if $x=1$ or $v=1$, which is problematic.

So you must put some vanishing condition on $G$ or use unbounded $L$, $R$.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.