Let $a_0$ and $b_0$ be smooth compactly supported functions in $B \subset R^3$, $f\in C^1(\Omega)$, and define

$a_n=f\Delta^{-1}(a_{n-1})=-f(x)\int_{B}a_{n-1}(y)\Phi(x-y)dy$, $n\geq 1$

$b_n=f\Delta^{-1}(b_{n-1})=-f(x)\int_{B}b_{n-1}(y)\Phi(x-y)dy$, $n\geq 1$

where $\Phi$ is the fundamental solution of the laplacian. Suppose

$\int_{B}(a_n-b_n)\varphi=0$ for every harmonic function $\varphi$ and every $n\geq 0$. Is $a_0=b_0$? When $f$ is a constant, the above means that $a_0-b_0$ is orthogonal to all functions $\varphi$ with $\Delta^n \varphi=0$ for some $n\in N$. This would imply $a_0=b_0$. Can this be generalized for a general coefficient $f$?

  • $\begingroup$ When $f\equiv1$, how does it imply $a_0=b_0$? $\endgroup$
    – timur
    Sep 21 '17 at 3:10
  • 1
    $\begingroup$ For every polynomial function $P(x,y,z)$ there exists an $n$ such that $\Delta^n P$ is harmonic. The proof will follow from an approximation argument. $\endgroup$
    – User4966
    Sep 21 '17 at 3:36
  • $\begingroup$ Just to be sure: is $\Phi(x) = |x|^{-1}$ the Newtonian potential? The Green function of $\Omega$ would be more natural here. $\endgroup$ Sep 23 '17 at 10:05
  • $\begingroup$ Yes $\Phi=|x|^{-1}$. $\endgroup$
    – User4966
    Sep 23 '17 at 15:05
  • $\begingroup$ Please do not make cosmetic edits to bump your question, given that you only asked it two days ago $\endgroup$
    – Yemon Choi
    Sep 23 '17 at 15:27

This is an extended comment, not an answer. Suppose that $a_0$ and $b_0$ satisfy the conditions given in the statement of the problem and $u := b_0-a_0$ is not identically zero.

The function $u$ is orthogonal to harmonic functions. Since the difference between $\Phi(x-y)$ and the Green function $G_\Omega(x,y)$ is a harmonic function, we have $\int_\Omega \Phi(x-y) u(y) dy = \int_\Omega G_\Omega(x,y) u(y) dy =: G_\Omega u(x)$. It follows that $f G_\Omega u$, $f G_\Omega(f G_\Omega u)$ and so on are orthogonal to harmonic functions.

If $f$ is positive and bounded, the operator $u \mapsto G_\Omega(f u)$, denoted by $G_\Omega f$ for simplicity, is a compact, non-negative definite, self-adjoint operator on $L^2(\Omega, f(x)dx)$, and hence it has an orthonormal basis of eigenfunctions $g_n$, with corresponding eigenvalues $\lambda_n$ converging monotonically to zero.

Orthogonality of $(f G_\Omega)^k u$ and a harmonic function $\phi$ is equivalent to orthogonality of $u$ and $(G_\Omega f)^k \phi$. Write $f^{-1} u$ and $\phi$ in the basis $(g_n)$: $f^{-1} u = \sum \alpha_k g_k$ and $\phi = \sum \beta_k g_k$. Then $$ 0 = \int_\Omega u(x) (G_\Omega f)^k \phi(x) = \sum \lambda_n^k \alpha_n \beta_n . $$ Let $M$ be the set of indices $m$ satisfying $\lambda_m = \max\{\lambda_n : \alpha_n \ne 0\}$. Multiplying both sides of the above equality by $\lambda_m^{-k}$ and passing to the limit as $k \to \infty$, we get $$ 0 = \sum_{m \in M} \alpha_n \beta_n . $$ Therefore, the function $v$ defined by $f^{-1} v = \sum_{m \in M} \alpha_n g_n$ is orthogonal to all harmonic functions. Furthermore, $f^{-1} v$ is the eigenfunction of $G_\Omega f$, that is, $G_\Omega v(x) = \lambda_m (f(x))^{-1} v(x)$.

This shows a different point of view on the original problem: take any function $v$ that is orthogonal to all harmonic functions, and evaluate $f(x) = v(x) / G_\Omega v(x)$. Can $f$ be bounded?

This is not an equivalent formulation, as it does not tell anything about the support of $u$ (which was required to be a compact subset of $\Omega$). But it seems to be reasonably close; in particular, if we can show that $v$ with the above properties does not exist, then no solution $u$ to the original problem can exist.

I bet $f$ cannot be bounded, but I have no proof. A heuristic argument: for $x$ close to a boundary point $x_0$ of $\Omega$, $G_\Omega(x, y)$ is close to a multiple of the Poisson kernel $P_\Omega(y, x_0)$, which is harmonic, and therefore orthogonal to $v$. Therefore, $G_\Omega v(x)$ should be of smaller order than $v(x)$ as $x$ converges to $x_0$, and so $f$ cannot be bounded.

Edited: I forgot to include two relatively simple remarks.

(1) If $\Omega$ has no holes (that is, its complement is connected), then it does not matter whether we assume that $a_n - b_n$ is orthogonal to harmonic functions on $\Omega$ or on $\mathbb{R}^3$. Indeed, by Runge's theorem for harmonic functions, for any compact subset $K$ of $\Omega$, any harmonic function on $\Omega$ can be approximated uniformly on $K$ by harmonic polynomials.

(2) If $f$ is zero on some ball $B$, then the answer is no: any pair $a_0, b_0$ such that $a_0 - b_0$ is orthogonal to harmonic functions and $a_0 - b_0 = 0$ outside $B$ has the desired conditions. Indeed: $\int_\Omega (a_0(y) - b_0(y)) \Phi(x - y) dy = 0$ outside $B$ because $\Phi(x - \cdot)$ is a harmonic function on (a neighbourhood of) $B$, and by remark (1), $a_0 - b_0$ is orthogonal to harmonic functions on (a neighbourhood of) $B$. This means that $a_1 = b_1$, and consequently $a_n = b_n$ for all $n \geqslant 1$. On the other hand, it is not difficult to construct a function $a_0 - b_0$ supported in $B$ and orthogonal to harmonic functions.


In general I think the conjecture is WRONG.but i do not find a counterexample.First of all,we can restate the conjecture in the frame of distribution theory:

$T$ is a distribution,iff $\forall g\in \Omega$,$Tg=0$,then $T$is the zero distribution,where $\Omega=\cup_{n\in N^*}\Omega_n$,$\Omega_n=\{g\in C_{c}^{\infty}(B),(f\Delta)^ng=0\}$.

It is easy to see if $f=constant$,then $T$ is a zero distribution.in fact we just need to check $\forall h\in D(B)$,which is the test function space,$T(h)=0$,because $T$ is a continuous linear functional on $D(B)$,so $h_n \longrightarrow h$ in$D(B)$ $\Longrightarrow$ $T(h)=\lim_{n\to \infty}T(h_n)$.and it is easy to see $\forall h\in D(B)$,there is a series of polynomial $h_n$,$h_n\longrightarrow h$in $D(B)$.and polynomial $h_n\in \Omega$.

this proof easy to generalize to the case when f is a linear function or some quadratic polynomial,because in this case we also have polynomial set $P\subset \Omega$.

but for f is a polynomial with $Deg(f)\geq3$,we always have $P-\Omega \neq \emptyset$.this can not derive the conjecture is wrong for $f$ because polynomial function is not a set of orthogonal basis.but this still give us some clue to given counterexamples.

on the other side,we can extend the set of $f$ to make the conjecture hold,for example consider $f$ to solve the equation $\Delta f=f$,then $\Omega=\{g|\exists n\in N^*,\sum_{k=1}^n \Delta^kg=0\}$.in this case I conjecture it is unessesary for $T$ to be zero.


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.