I have two questions about inhomogenous Fredholm integral equations of the first kind: $$f(x) = \int_a^b K(x,t) g(t) dt$$ where $f, K$ are known and $g$ is not.
If a unique solution for $g$ exists, under what conditions is $g$ a probability density function -- i.e., a positive function that integrates to 1?
For the regression equation $$r(x) \equiv E[Y | X=x] = \int_a^b E[Y | X=x, Z=z; \beta] g(z) dz,$$ solved by $g$ and $\beta$, must there exist another solution with $g’\neq g$ and $\beta’\neq\beta$? The regression function integrates over $Z$, which is not observed. I am particularly interested in cases of the form $$r(x) =\int_a^b \exp(-\exp(z + \beta x))\,g(z)dz$$ and using this to show that the parameter $\beta$ is not statistically identified.