note 6 : is curvelet a good setting for pqn

[LinaMiao]

Since high dimension(bigger than one) curelet are redundant,

say one example I have been used in curve let case,

whos x_spg
Name Size Bytes Class Attributes

x_spg 529339x1 4234712 double

whos A
Name Size Bytes Class Attributes

A 145408x529339 210093 opFoG

whos RM
Name Size Bytes Class Attributes

RM 145408x182272 206777 opKron

whos D
Name Size Bytes Class Attributes

D 182272x1 1458176 double

Which means I have to store several vectors of the length 529339,

Well if I'm not using a redundant dictionary, say DFT,

F = 182272 _182272
A = 145408_182272
x = 182272*1

But, this makes the matrix small, which result in a cheaper mat-vec multiplication,

So, where is the balance