Ngô Quốc Anh

May 6, 2011

Laplace-Beltrami vs. Laplacian

Filed under: Riemannian geometry — Ngô Quốc Anh @ 8:10

Let us do compare $\Delta_g$ with respect to metric $g$ and $\Delta$ with respect to Euclidean metric. For simplicity, let us try with function $u^{-\beta}$. Obviously, $\displaystyle\begin{gathered} \Delta ({u^{ - \beta }}) = {\partial ^i}{\partial _i}({u^{ - \beta }}) \hfill \\ \qquad= {\partial ^i}( - \beta {u^{ - \beta - 1}}{\partial _i}u) \hfill \\ \qquad= \beta (\beta + 1){u^{ - \beta - 2}}({\partial ^i}u)({\partial _i}u) - \beta {u^{ - \beta - 1}}{\partial ^i}{\partial _i}u \hfill \\ \qquad= \beta (\beta + 1){u^{ - \beta - 2}}|\nabla u{|^2} - \beta {u^{ - \beta - 1}}\Delta u \hfill \\ \qquad= - \beta {u^{ - \beta - 1}}\left( {\Delta u - (\beta + 1)\frac{{|\nabla u{|^2}}}{u}} \right). \hfill \\ \end{gathered}$

Next, by definition of $\Delta_g$ we obtain $\displaystyle\begin{gathered} {\Delta _g}({u^{ - \beta }}) = \frac{1}{{\sqrt {\det g} }}{\partial _j}\left( {\sqrt {\det g} {g^{ij}}{\partial _i}({u^{ - \beta }})} \right) \hfill \\ \qquad= \frac{1}{{\sqrt {\det g} }}{\partial _j}\left( {( - \beta ){u^{ - \beta - 1}}\sqrt {\det g} {g^{ij}}{\partial _i}u} \right) \hfill \\ \qquad= \frac{1}{{\sqrt {\det g} }}\left[ {\beta (\beta + 1){u^{ - \beta - 2}}{\partial _ju}\sqrt {\det g} {g^{ij}}{\partial _i}u - \beta {u^{ - \beta - 1}}{\partial _j}\left( {\sqrt {\det g} {g^{ij}}{\partial _i}u} \right)} \right] \hfill \\ \qquad= \beta (\beta + 1){u^{ - \beta - 2}}|\nabla u{|_g^2} - \beta {u^{ - \beta - 1}}\Delta u_g \hfill \\ \qquad= - \beta {u^{ - \beta - 1}}\left( {\Delta_g u - (\beta + 1)\frac{{|\nabla u{|_g^2}}}{u}} \right). \hfill \\ \end{gathered}$

Thus $\displaystyle {\Delta _g}({u^{ - \beta }}) = \Delta ({u^{ - \beta }}).$

So the question is why do they equal?

1. You can just put $\beta=-1$ in the second computation to see what’s wrong. I think you want to compare the laplacian of functions in Euclidean space under different metrics right? If we denote the laplacian with respect to $g$ by $\Delta_g$ and the corresponding gradient by $\nabla_g$, then your computation should read $\displaystyle\Delta_g(u^{-\beta})=\beta(\beta+1)u^{-\beta-2}|\nabla_g u|_g^2-\beta u^{-\beta-1}\Delta_g u.$ $|\nabla_g u|_g^2$ etc. are, of course, not the same under different $g$.

Comment by K — June 1, 2011 @ 21:45

• Ah yes, thank for your kind explanation. So my computation was correct but $\displaystyle {\Delta _g}({u^{ - \beta }}) \ne \Delta ({u^{ - \beta }})$

since the gradient and Laplacian depend strongly on metric. I hope you can continue contributing your opinion here.

Comment by Ngô Quốc Anh — June 1, 2011 @ 22:11

This site uses Akismet to reduce spam. Learn how your comment data is processed.