IsotopeQuantity.decays_from and IsotopeQuantity.from_decays can suffer in precision if halflife >> time interval. This is because the calculations involve a 1 - np.exp(-decay_const * time) kind of expression, and if decay_const * time is very small, the exponent is very close to 1, and the difference loses precision because of these two floating point quantities. See np.spacing.
Better precision could be achieved in some cases by a Taylor series approximation:
The code could use np.spacing to check whether the series approximation or floating point has better precision.
This is relevant for long-lived isotopes like K-40. (I'm getting ~0.1% loss of precision for an hour's interval with K-40, which was causing tests to fail.)
jvavrek
commented
2 years ago
IsotopeQuantity.decays_fromandIsotopeQuantity.from_decayscan suffer in precision if halflife >> time interval. This is because the calculations involve a1 - np.exp(-decay_const * time)kind of expression, and ifdecay_const * timeis very small, the exponent is very close to 1, and the difference loses precision because of these two floating point quantities. Seenp.spacing.Better precision could be achieved in some cases by a Taylor series approximation:
The code could use
np.spacingto check whether the series approximation or floating point has better precision.This is relevant for long-lived isotopes like K-40. (I'm getting ~0.1% loss of precision for an hour's interval with K-40, which was causing tests to fail.)