Questions about PRIME_BITLEN and round number

[XuyangSong]

Plonk-Prototype/src/poseidon/round_numbers.rs

Line 7 in bae3f95

const PRIME_BITLEN: usize = 256;

Can we always use PRIME_BITLEN=256 to compute round number for different fields?

Another question about round number
Does the round number generated in the repo(rust) still have the same result with reference from the python script here?
I don't know why I could not get the same result.

Btw, for generate_parameters_grain_deterministic.sage and poseidonperm_bls381_width3.sage, I can get the same result with rust code even for different fields.

[BoyuanFeng]

1. "Can we always use PRIME_BITLEN=256 to compute round numbers for different fields?"

No, you cannot. As specified in the comments (see below), PRIME_BITLEN=256 is "the number of bits of the Poseidon prime field modulus."

Plonk-Prototype/src/poseidon/round_numbers.rs

Lines 3 to 6 in bae3f95

	// The number of bits of the Poseidon prime field modulus. Denoted `n` in the Poseidon paper
	// (where `n = ceil(log2(p))`). Note that BLS12-381's scalar field modulus is 255 bits, however we
	// use 256 bits for simplicity when operating on bytes as the single bit difference does not affect
	// the round number security properties.

2. "Does the round number generated in the repo(rust) still have the same result with reference from the python script here?"

Yes. For example, if you run "python2 calc_round_numbers.py 762 3 128 1 1", you will have (R_F, R_P) = (8, 55). This matches hardcoded tests:

Plonk-Prototype/src/poseidon/round_numbers.rs

Line 101 in bae3f95

(3, 55),

[XuyangSong]

Thanks for your reply.

1. So we'd better not hardcode the PRIME_BITLEN , and use Fr: : size_in_bits() instead or take another parameter?
2. Oh, I used the wrong sbox_case parameter.

[BoyuanFeng]

So we'd better not hardcode the PRIME_BITLEN , and use Fr: : size_in_bits() instead or take another parameter?

It is indeed more complex than that. The properties used to compute round numbers also change from the current implementation. You may also need to update alpha. Please check the paper for properly selecting round numbers for other curves.

[XuyangSong]

ty, I'll check it.