Author:
Gariboldi Bianca,Gigante Giacomo
Abstract
AbstractWe show how to build a kernel $$ K_X(x,y)=\sum _{m=0}^Xh(\lambda _m/{\lambda _X})\varphi _m(x)\overline{\varphi _m(y)} $$
K
X
(
x
,
y
)
=
∑
m
=
0
X
h
(
λ
m
/
λ
X
)
φ
m
(
x
)
φ
m
(
y
)
¯
on a compact Riemannian manifold $${{\,\mathrm{\mathcal {M}}\,}}$$
M
, which is positive up to a negligible error and such that $$K_X(x,x)\approx X$$
K
X
(
x
,
x
)
≈
X
. Here $$0=\lambda _0^2\le \lambda _1^2\le \cdots $$
0
=
λ
0
2
≤
λ
1
2
≤
⋯
are the eigenvalues of the Laplace–Beltrami operator on $${{\,\mathrm{\mathcal {M}}\,}}$$
M
, listed with repetitions, and $$\varphi _0,\,\varphi _1,\ldots $$
φ
0
,
φ
1
,
…
an associated system of eigenfunctions, forming an orthonormal basis of $$L^2({{\,\mathrm{\mathcal {M}}\,}})$$
L
2
(
M
)
. The function h is smooth up to a certain minimal degree, even, compactly supported in $$[-1,1]$$
[
-
1
,
1
]
with $$h(0)=1$$
h
(
0
)
=
1
, and $$K_X(x,y)$$
K
X
(
x
,
y
)
turns out to be an approximation to the identity.
Funder
Università degli studi di Bergamo
Publisher
Springer Science and Business Media LLC