I use high value dimension of inputs vector for multiplication and depth=1(python wrapper)
like
x1 = [2**10+0.251, 2**10+0.0251, 2**10 + 0.251, 2**10 + 0.0251, 2**10+0.251, 2**10+0.0251, 2**10+0.251, 2**10+ 0.0251]
x2=x1
After encryption i get something like overflow:
x1 * x2 = (514.112, 51.4065, 514.112, 51.4065, 514.112, 51.4065, 514.112, 51.4065, ... ); Estimated precision: 32 bits
Error is 2**20
Why do i not get approxation error. What is expalanation about this behavior?
You get an overflow because the room between FirstModSize (60 by default) and ScaleModSize is too small to fit the results. A simple way to fix it is to set the depth to 2 (it will increase the room by extra 60 bits).
Is there a way/formel how can i calculate the range of input vectors to avoid overflow and/or approximation error?(issue about ckks)
I suggest reading the initial sections of https://eprint.iacr.org/2020/1118