Consider a probability measure $d\mu = w(t) dt$ with $w(t)\in L^1(I)$, $I =\left[ 0,1\right]$. What are the minimal assumption I can take on two functions $f,g:I\ \to \mathbb{R}$ so that an integration by parts -like statement could be made, and what would it be? Also, can the demands on measure space $\left(\Omega, \mu\right)$ be relaxed so that this result would be still valid? Something with non compact $\Omega$, infinite measure etc. [This post][1] asks a somehow related question, but not quite the same question. [1]: http://mathoverflow.net/questions/119407/integration-by-parts-formula-for-functionals Thanks Amir