1547845440-The_Ricci_Flow_-_Techniques_and_Applications_-_Part_III__Chow_

(jair2018) #1

  1. EXISTENCE OF THE HEAT KERNEL FOR A TIME-DEPENDENT METRIC 273


2.2. The parametrix convolution series.
Similarly to the previous chapter, we shall construct the heat kernel by

a convolution series. Recall that IRS, = { ( T, v) E IR^2 : T > v}.


DEFINITION 24.13 (Space-time convolution). Given two functions
H, J E c^0 (M x M x IRS,),

their space-time convolution is defined by


(24.28) (H * J) (x, T; y, v) ~ 1


7
JM H (x, T; z, (}) J (z, (}; y, v) dμg(CT) (z) d(}

as long as the integral is well defined.
REMARK 24.14. Note that when g ((}) = g, by letting u = 0 and by tak-
ing H (x, t; z, s) = F (x, z, t - s) and J (z, s; y, u) = G (z, y, s - u) in (24.28),
we have


(H * J) (x, t; y, u) ~lat JM F (x, z, t - s) G (z, y, s) dμ 9 (z) ds,


which is the same as (23.46).


LEMMA 24.15. The space-time convolution in (24.28) is associative.
PROOF. We compute
(24.29)

((H J) K) (x, T; y, v)


= 1


7
JM (H * J) (x, T; z, (}) K (z, (}; y, v) dμg(CT) (z) d(}

=1T1T r H(x, T; w, p) J(w, p; z, (}) K(z, (}; y, v)dμg(p)(w)dμg(CT)(z)dpd(}
v CTJMxM ·

and


( H (J K)) ( x' T; y' v)


= 1


7
JM H (x, T; z, (}) (J * K) (z, (}; y, v) dμg(CT) (z) d(}

=1T1CT r H(x, T; z, (}) J(z, (}; w, p) K(w, p; y, v)dμg(p)(w)dμg(CT)(z)dpd(}.


v vJMxM


Exchanging the order of integration with respect to p and (} (and switching
the labels w and z), we obtain


(H (JK)) (x,T;y,v)


=1


7
r r H(x, T; w, (}) J(w, (}; z, p) K(z, p; y, v)dμg(p)(z)dμg(CT)(w)d(}dp
v }p JMxM

=1T1T r H(x, T; w, p) J(w, p; z, (}) K(z, (}; y, v)dμg(CT)(z)dμg(p)(w)dpd(},
v CT JMxM
where in the last line we relabelled p and(}"; this is the same as (24.29). D
Free download pdf