math: support power operation with two real numbers (std::Decimal)

[byeongsu-hong]

Is there any plans to support power operation between two Decimals? I'm currently developing the pool swap simulator on Osmosis, but there's some limitation of implementing "math stuffs" like pow operation.

[webmaster128]

Which operation exactly do you need?

[byeongsu-hong]

Which operation exactly do you need?

I need operation between Bt and Wt in a real number

[webmaster128]

I would not recommend doing that kind of math with cosmwasm_std::Decimal or cosmwasm_std::Decimal256 because they are both fixed point decimals. This means especially multiplications and powers can quickly lead to significant rounding errors. I'd look for a library specific for that kind of job and have a layer of converting inputs and outputs.

Real numbers could be implemented with either floats or floating point decimals.

Floats are currently disabled, but might be shipped as part of the next release if everything goes well: #1845. If that works out you can use native f32/f64.

Floating point decimals are like Decimal but the dot is moving instead of being always 18 positions left of the end. This allows very high values or very low values with much lower loss of precision. Some examples for that approach are:

• rust-decimal which has impl Pow<Decimal> for Decimal (not used myself but work for Proof of Engagement)
• DEC64: Decimal Floating Point
• ... probably more, did not research much

[byeongsu-hong]

rust-decimal which has impl Pow for Decimal (not used myself but work for Proof of Engagement)

I'm using this library to solve this problem, but it's not fit with the big numbers from Osmosis' weighted pool. I think it does not solve the issue that i surfaced even if cosmwasm supports f32/f64.

This means especially multiplications and powers can quickly lead to significant rounding errors

I'm totally agree with your opinion and this should not be applied on this library for same reason.

I found Osmosis team uses Taylor Expansion + big.Int to approximate pow operation. So there's two point to choose if i really want to use pow operation.

1. Wait until Osmosis team opens query for using their math.
2. Make my own math library with Uint256 or something

[webmaster128]

What's a typical value range for those t, B and W? Maybe the 18 digit fixed point is good enough?

[byeongsu-hong]

I'll let you know if i check the actual value of it

[chipshort]

Not planned for now.