Timeline for How to add large exponential terms reliably without overflow errors?
Current License: CC BY-SA 3.0
11 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Nov 4, 2017 at 1:43 | comment | added | emsr | try sorting the a_i and accumulate the smallest terms first. That way they won't be lost when added to large terms. | |
| Aug 4, 2016 at 19:39 | comment | added | a06e | scicomp.stackexchange.com/q/24624/988 | |
| Aug 4, 2016 at 17:43 | comment | added | a06e | what if you have many very small terms? It could happen that $e^{a_i - K} \approx 0$ for these. If there are many terms like this, you would have a large error. | |
| Feb 13, 2012 at 0:06 | comment | added | Jack Poulson | To the downvoter: would you mind letting me know what is wrong with my answer? | |
| Feb 3, 2012 at 16:59 | vote | accept | cboettig | ||
| Feb 2, 2012 at 0:57 | comment | added | Jack Poulson | Ah, I now see what you were getting at. You actually don't need to worry about underflow, as adding exceptionally tiny results to your solution shouldn't change it. If there was an exceptionally large number of them, then you should sum the small values first. | |
| Feb 2, 2012 at 0:53 | comment | added | cboettig | Thanks for the clear notation -- but I believe this is essentially what I have proposed(?) If I need to avoid underflow errors when some $a_i$ are small, I gather I need the Kahan summation approach proposed by @gareth? | |
| Feb 2, 2012 at 0:29 | history | edited | Jack Poulson | CC BY-SA 3.0 |
rewording first sentence
|
| Feb 1, 2012 at 23:45 | history | edited | Geoff Oxberry | CC BY-SA 3.0 |
Replaced Jack's "a" with "K" for consistency with cboettig's notation.
|
| Feb 1, 2012 at 21:00 | history | edited | Jack Poulson | CC BY-SA 3.0 |
added 23 characters in body
|
| Feb 1, 2012 at 20:47 | history | answered | Jack Poulson | CC BY-SA 3.0 |