Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

Suppose $W : \Bbb{R}^n \to \Bbb{R}_+$ is a continuous, positive function, with exactly $n$ zeros $\alpha_1,...,\alpha_n$. Define the following 'distance':

$$ d(\alpha_i,\alpha_j)=\inf\{\int_0^1 \sqrt{W(\gamma(t))}| \gamma'(t)|dt : \gamma \in C^1([0,1];\Bbb{R}^n), \gamma(0)=\alpha_i,\ \gamma(1)=\alpha_j\}$$

Suppose I have a set of real, positive numbers $\sigma_{ij}>0,\ i \neq j$ with the property that $\sigma_{ij}=\sigma_{ji}$ and $\sigma_{ij} \leq \sigma_{ik}+\sigma_{kj},i,j,k=1,...,n$.

My question is:

Can we find $\alpha_i, i=1..n$ and $W$ with the desired properties, such that $d(\alpha_i,\alpha_j)=\sigma_{ij}$?

I feel that the fact that we can choose $W$ and the zeros of $W$, $\alpha_1,...,\alpha_n$ gives enough freedom for us to solve this system. Thank you.


I should say why I need to know if this result is true. I am studying an article of Baldo: Minimal interface criterion for phase transitions in mixtures of Cahn-Hilliard fluids, where he proves that the following functional $$ \mathcal{F}(E_1,...,E_n) = \sum_{1\leq i < j\leq n} d(\alpha_i,\alpha_j) \mathcal{H}^{N-1}(\partial^*E_i \cap \partial^*E_j \cap \Omega) $$ is a $\Gamma$-limit of certain functionals, and therefore it is lower semicontinuous, where $d(\alpha_i,\alpha_j)$ is defined as above. I was wondering if it is possible to prove that for any $\sigma_{ij}$ which satisfy the triangle inequality (which is a necessary physical condition), the lower semicontinuity, and therefore the existence of a minimum for the given energy, $$ \mathcal{F}(E_1,...,E_n) = \sum_{1\leq i < j\leq n} \sigma_{ij} \mathcal{H}^{N-1}(\partial^*E_i \cap \partial^*E_j \cap \Omega) $$ still holds.

share|improve this question
add comment

1 Answer

Your question is somewhat poorly phrased: The condition $\gamma>0$ in the definition of the "distances" $d(\alpha_i,\alpha_j)$ is meaningless and should be dropped. Once you do this, the object you are looking at is a singular conformally-Euclidean Riemannian metric $g= \rho(x) g_0$, where $g_0$ is the flat metric on ${\mathbb R}^n$ and $\rho(x)=W^2(x)$ is a continuous function which (for some unexplained reasons) is required to vanish at the points $\alpha_1,...,\alpha_n$. For such singular metrics one defines the usual notion of length of a piecewise-smooth path and distance function $d(x,y)$ (by infimizing lengths of paths). Then your numbers $d(\alpha_i,\alpha_j)$ are nothing but the distances between the singular points $\alpha_i,\alpha_j$. Thus, a better title of your question is probably "Singular conformally-Euclidean metrics" and right tag is differential geometry.

Now, your question has positive but not pretty answer. Furthermore, you can even do this with $n$ prescribed singular points in ${\mathbb R}^2$. Here is a sketch (in ${\mathbb R}^n$ as you requested). Start with $n$ points $\alpha_1,...,\alpha_n$ in ${\mathbb R}^n$ within unit distance from each other. Let $\Gamma$ be the complete graph that these points span (connect points by straight-line segments). On each segment $E_{ij}:=[\alpha_i, \alpha_j]$ choose a piecewise-linear function $W_{ij}$ (with one break-point, say, at the mid-point $m_{ij}$ of $E_{ij}$) which vanishes at the end-points and whose integral over $E_{ij}$ equals $\sigma_{ij}$. Thus, we defined a singular Riemannian metric on the graph $\Gamma$ and the corresponding distance function. The triangle inequalities that you impose on the numbers $\sigma_{ij}$ will ensure that the distances between points $\alpha_i, \alpha_j$ on $\Gamma$ are exactly $\sigma_{ij}$. Next, you will extend $W$ to small conical neighborhoods $C_{ij}$ of the segments $E_{ij}$, where each $C_{ij}$ is the union of two cones of revolution with vertices $\alpha_i, \alpha_j$, small vertex angles, axes $[\alpha_i, m_{ij}], [\alpha_j, m_{ij}]$ and equal bases orthogonal to $E_{ij}$. Let $C$ denote the union of all the "double cones" $C_{ij}$. Let $p_{ij}$ denote the orthogonal projection of $C_{ij}$ to $E_{ij}$. Extend $W_{ij}$ to ${\mathbb R}^n$ by the formula: $W_{ij}(x)=(1+ c d_{ij}(x)) W_{ij}(p_{ij}(x))$. Here $d_{ij}(x)$ is the distance from $x$ to $E_{ij}$ and $c\ge 0$ is a constant. Now, restrict $W_{ij}$ to $C_{ij}$ and let $W$ denote the resulting function on $C$. Then it is easy to see that for every choice of $c\ge 0$, for the singular Riemannian metric $g=W^2 g_0$ on $C$, the distances between the points $\alpha_i, \alpha_j$ are exactly $\sigma_{ij}$. (Use the fact that each $p_{ij}$ will contract the metric $g$.)

Lastly, you have to extend the function $W$ to the rest of ${\mathbb R}^n$. This is where things will become ugly especially near the points $\alpha_i$. It will be up to you to do the dirty part (assuming that you really need it). If $\epsilon>0$ is given and $c$ is large enough, then outside of Euclidean $\epsilon$-neighborhoods of the points $\alpha_k$ you can take $W$ equal $$ W(x)= \min_{i\ne j} W_{ij}(x) $$ where $W_{ij}$ are defined on the entire ${\mathbb R}^n$ as above. The hard part is taking care of Euclidean $\epsilon$-balls centered at the points $\alpha_k$. It might be easier to do the extension if you allow each piecewise-linear function $W_{ij}$ to have two break-points so that slopes of $W_{ij}$ (on $E_{ij}$) are the same at all points $\alpha_k$.

share|improve this answer
    
Thank you very much for your guidelines and for your answer. I edited my question using your suggestions, and I will say why I needed to know if this result holds in a next edit of my question. –  Beni Bogosel Mar 2 '12 at 10:39
add comment

Your Answer

 
discard

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.