28.12.2014 Views

Extremum problems for eigenvalues of discrete Laplace operators

Extremum problems for eigenvalues of discrete Laplace operators

Extremum problems for eigenvalues of discrete Laplace operators

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

10 REN GUO<br />

Then 0 = ∂H<br />

∂θ 1<br />

and 0 = ∂H<br />

∂θ 2<br />

imply that<br />

∑ m<br />

∑ m<br />

(a 3 ...a m + a 2 a 3 ...â i ...a m )(1 + a 2 1) = (a 3 ...a m + a 1 a 3 ...â i ...a m )(1 + a 2 2).<br />

i=3<br />

Since a 1 + a 2 > 0, it is equivalent to<br />

(a 1 − a 2 )(a 1 + a 2 )(a 3 ...a m + a 1a 2 − 1<br />

a 1 + a 2<br />

The third factor is<br />

cot θ 3 ...cot θ m + cot(θ 1 + θ 2 )<br />

m<br />

∑<br />

i=3<br />

i=3<br />

a 3 ...â i ...a m ) = 0.<br />

m∑<br />

cot θ 3 ...ĉot θ i ...cot θ m<br />

which is written as ∑ m<br />

i=1 ã1...̂ã i ...ã m−1 , where ã 1 = cot(θ 1 + θ 2 ),ã i = cot θ i+1 <strong>for</strong><br />

i = 2,...,m−1. This expression corresponds to a cyclic (m−1)-gon. By assumption<br />

<strong>of</strong> the induction, ∑ m<br />

i=1 ã1...̂ã i ...ã m−1 > 0.<br />

Hence the only possibility is a 1 = a 2 . By similar argument, we show that a i = a j<br />

<strong>for</strong> any i,j. Hence the function ∑ m<br />

i=1 a 1...â i ...a m has the unique critical point such<br />

that θ i = π m<br />

<strong>for</strong> any i = 1,...,m.<br />

Next, we claim that ∑ m<br />

i=1 a 1...â i ...a m > 0. Without loss <strong>of</strong> generality, we assume<br />

that a 1 > 0,a 2 > 0,...,a m−1 > 0. Now<br />

m∑<br />

a 1 ...â i ...a m<br />

i=1<br />

i=3<br />

m−2<br />

∑<br />

= a 1 a 2 ...a m−2 (a m−1 + a m ) + a 1 ...â i ...a m−2 (a m−1 a m )<br />

i=1<br />

m−2<br />

∑<br />

m−2<br />

∑<br />

= a 1 a 2 ...a m−2 (a m−1 + a m ) + a 1 ...â i ...a m−2 (a m−1 a m − 1) + a 1 ...â i ...a m−2<br />

i=1<br />

m−2<br />

∑<br />

m−2<br />

a m−1 a m − 1 ∑<br />

= (a m−1 + a m )(a 1 a 2 ...a m−2 + a 1 ...â i ...a m−2 ) + a 1 ...â i ...a m−2 .<br />

a m−1 + a m<br />

Let ã m−1 = am−1am−1<br />

a m−1+a m<br />

i=1<br />

= cot(θ m−1 + θ m ). Then<br />

m−2<br />

∑ a m−1 a m − 1<br />

a 1 a 2 ...a m−2 + a 1 ...â i ...a m−2<br />

a m−1 + a m<br />

i=1<br />

m−2<br />

∑<br />

= a 1 a 2 ...a m−2 + a 1 ...â i ...a m−2 ã m−1 .<br />

i=1<br />

i=1<br />

i=1<br />

Consider an cyclic (m −1)-gon with angles θ 1 ,...,θ m−2 ,θ m−1 +θ m . By the assumption<br />

<strong>of</strong> induction,<br />

There<strong>for</strong>e ∑ m<br />

i=1 a 1...â i ...a m > 0.<br />

m−2<br />

∑<br />

a 1 a 2 ...a m−2 + a 1 ...â i ...a m−2 ã m−1 > 0.<br />

i=1

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!