Let's Talk About Consensus Dangers in Block Rewards

[kauffj]

It struck me that a fully continuous block reward function leveraging exponential decay would possibly be superior to many other algorithms. Then it struck me that such algorithms may face trouble reaching consensus, due to differing floating point math in different architectures.

Here is an example of the type of formula I am talking about:

http://lbry.io/img/block-formula-proposed.png

Or as text:

F(n) = c * (4000000 - n) ^ e / 4000000

Where F(n) is the block reward, n is the block number, c is a scalar constant, and e is Euler's number.

Would this type of formula risk differing architectures disagreeing on the block reward due to different strategies on floating point math and rounding? Or is there a way to design an algorithm similar to this one that does not face these risks? Or am I being overly crazy and this problem does not exist whatsoever?

Any thoughts or comments would be appreciated. Any references that show other coins using or dismissing this type of algorithm doubly so.

[1715484908]

1715484908

[1715484908]

1715484908

[1715484908]

1715484908

[monsterer]

You'd have to use an integer math approximation; reward can only be 1 satoshi minimum anyway, never a fractional amount, so floating precision may not be necessary.