Positive definiteness of graph Laplacian

90 Views Asked by At

Consider the proof at page 2 found here: https://people.orie.cornell.edu/dpw/orie6334/Fall2016/lecture7.pdf

I can’t wrap my head around the second and third line:

\begin{align} &= \sum_{i \in V}x(i)^2 - \sum_{(i, j) \in E}\frac{2x(i)x(j)}{\sqrt{d(i)d(j)}}\\ &= \sum_{(i, j) \in E}\left(\frac{x(i)}{\sqrt{d(i)}} - \frac{x(j)}{\sqrt{d(j)}}\right)^2 \end{align}

They say that it can be seen by “inversely” expanding the second term. However, when I do so, to me at least it is definitely not obvious: $$\left(\frac{x(i)}{\sqrt{d(i)}} - \frac{x(j)}{\sqrt{d(j)}}\right)^2 = \frac{x(i)^2}{d(i)} - \frac{2x(i)x(j)}{\sqrt{d(i)d(j)}} + \frac{x(j)^2}{d(j)}.$$

Now from this I have no clue how this can be related to the first line.

How did they go from the first line above to the second, what are the intermediate steps?

1

There are 1 best solutions below

0
On

The notation is somewhat confusing in that the edges are denoted $(i,j)\in E$ as if they were ordered pairs, but the graph is actually undirected and the sum is really meant to go over each undirected edge once. Thinking about each edge as a two-element set, the argument is just exchanging the order of summation: $$\begin{align*} \sum_{\{i,j\}\in E}\frac{x(i)^2}{d(i)}+\frac{x(j)^2}{d(j)}&= \sum_{e\in E}\sum_{i\in e}\frac{x(i)^2}{d(i)}\\ &=\sum_{i\in V}\sum_{\substack{e\in E\\i\in e}}\frac{x(i)^2}{d(i)}\\ &=\sum_{i\in V}\frac{x(i)^2}{d(i)}\sum_{\substack{e\in E\\i\in e}}1\\ &=\sum_{i\in V}\frac{x(i)^2}{d(i)}d(i)\\ &=\sum_{i\in V}x(i)^2. \end{align*}$$