Core idea

Let uu be the of sufficiently regular boundary data ff on R\mathbb R. The Dirichlet-to-Neumann operator sends ff to the boundary normal derivative

Λf=yu(+i0).\Lambda f=-\partial_yu(\,\cdot\,+i0).

With this outward-normal convention,

Λf=H(f)=(x2)1/2f,\Lambda f=H(f')=(-\partial_x^2)^{1/2}f,

where HH is the .

Fourier symbol

The Poisson extension satisfies u^(ξ,y)=e2πξyf^(ξ)\widehat u(\xi,y)=e^{-2\pi|\xi|y}\widehat f(\xi). Hence Λf^(ξ)=2πξf^(ξ)\widehat{\Lambda f}(\xi)=2\pi|\xi|\widehat f(\xi) for the 2π2\pi Fourier normalization.

Sign convention

Using the inward derivative +y+\partial_y replaces Λ\Lambda by Λ-\Lambda. Statements involving (H[-f']) use that alternative convention.

References
  1. Luis Caffarelli and Luis Silvestre, “An extension problem related to the fractional Laplacian,” Communications in PDE 32 (2007), 1245–1260. DOI record.