Stable log-sum-exp on overflowing inputs
Compare naive exponentiation with a shifted log-sum-exp calculation and a 70-digit Decimal reference on three fixed large inputs.
Proposed original example. Source release and any execution are separate steps.
Can a shifted calculation retain an accurate finite result when direct exponentiation overflows?
Compare the shifted float result with a separately formulated high-precision Decimal sum and logarithm; inspect overflow as a distinct outcome.
What you could produce
- A scoped observations JSON record and a comparison with the stated reference.
Before you use it
- Python 3.13 standard library
Limits to keep in view
- No research code was executed by the preparation tool.
- Author output cannot issue an independent scientific-verification result.
Source and permission context
Original local preparation by the Executable Science seed collection; upstream API references remain separately attributed.
Rights need review. Review the scope and upstream conditions before reuse.
Still unresolved
- Proposed local-draft licenses: original code MIT, explanations CC-BY-4.0, synthetic numeric data CC0-1.0; publication/disclosure approval remains separate.