Notation. $H^k(\mathbb R^n)$ denotes the Sobolev space $W^{k,2}(\mathbb R^n)$. The Sobolev space $W^{k,p}(\mathbb R^n)$ is the Banach space of functions in $L^p$ with all derivatives up to order $k$ in $L^p$.
My question is about non-uniqueness of weak solutions of an elliptic PDE.
The PDE
Consider a very simple semilinear elliptic PDE: $$ \left\{\begin{aligned} &-\Delta u=\partial_{x_1}(u^2+f)& &\text{in } \mathbb R^n,\\ & \,\,\,u(x)\to 0\,,& &x\to\infty. \end{aligned}\right. $$ where $x=(x_1,...,x_n)$. The boundary conditions are not rigorous, it is just to describe the kind of solutions we want. The critical Lebesgue spaces with respect to the natural scaling symmetry are: $$ u\in L^{n}(\mathbb R^n), f\in L^{n/2}(\mathbb R^n). $$
Strong solutions
Assume $n=3$ to fix ideas. By using the fixed point theorem, it is possible to show that if $f\in L^{3/2}(\mathbb R^3)$ is small enough, then there exists a unique small solution $u\in L^3(\mathbb R^3)$, which moreover satisfies $$ \|\nabla u\|_{L^{3/2}(\mathbb R^3)}\lesssim \|f\|_{L^{3/2}(\mathbb R^3)} $$ (note that $\dot W^{1,3/2}(\mathbb R^3)\hookrightarrow L^3(\mathbb R^3)$).
Weak solutions
If I instead consider a scaling-subcritical datum, like $f\in L^2(\mathbb R^3)$, I cannot use the fixed point theorem. But notice that multiplying the equation by $u$ and integrating by parts, we obtain the a priori bound
$$ \|\nabla u\|_{L^2(\mathbb R^3)}\lesssim \|f\|_{L^2(\mathbb R^3)}. $$
Using this a-priori bound (please correct me if I am wrong*) it is possible to prove the existence of weak solutions $u\in \dot H^1(\mathbb R^3)$ for any $f\in L^2(\mathbb R^3)$ (we have $\dot H^1(\mathbb R^3)\hookrightarrow L^6(\mathbb R^3)$).
In this case, estimates for the difference of two solutions do not seem to exist, so the uniqueness cannot be proved by standard arguments (this is the only reason why I am calling them “weak solutions”; you can see that they are more regular than strong solutions).
My question
My problem is the following. Is the weak solution unique if $f\in L^2(\mathbb R^3)$? If no, how can one find a counterexample?
In a case where we have a priori bounds like in the PDE I wrote, do we have non-uniqueness in case of subcritical data? Note that this is not a regularity prolem: data in $L^2$ are "more regular" than data in $L^{3/2}$. The problem is that data grow more slowly at infinity than data in the critical space. I could ask the same question for data $f\in H^{100}(\mathbb R^3)$, and the answer would still be non trivial. Also, I am reasonably sure that a unique continuation property exists for this kind of equation, so this problem is extremely different from those of non-uniqueness of weak solutions of evolution PDE at low regularity (like, e.g., the famous results for Euler equation).
I feel it is likely that there are data that lead to two different solutions, but I know no technique that would do that or examples like that in the literature. If you know of articles about anything that resembles this problem, any reference is welcome.
*I think there should be more than one way of doing that. The way I thought would be to approximate the Cauchy problem with that of the same PDE on a half plane (with Dirichlet boundary conditions), where the "edge" of the plane goes to infinity. On a half plane, you can look at the PDE as an evolution equation via the Poisson kernel, so you can show that a solution exists, and then you can use the uniform bound and weak compactness to show the existence of the solution of the original problem... Probably not the easiest way. I have to say I am not an elliptic person, I am mostly dispersive, sometimes parabolic... In any case, feel free to comment on this point.