R.PGUMBEL Function (Gnumeric)
Calculate Gumbel lower and upper probabilities with Gnumeric R.PGUMBEL. Verify location/scale meaning, log tails and an extreme upper-log-tail underflow limit.
Verification status: 20 reference, boundary and advanced-check formulas were recalculated with Gnumeric 1.12.57 ssconvert. Saved identities, the blank input and numerical results were independently checked. Gnumeric 1.12.62 was source-checked only. How formula examples are checked
Extreme upper-log-tail limitation: R.PGUMBEL(1000,0,1,FALSE(),TRUE()) returned #NUM! in Gnumeric 1.12.57. Its mathematical log survival probability is finite and approximately −1000; an earlier exponential intermediate underflows in the inspected helper.
Quick answer: =R.PGUMBEL(0,0,1) gives 0.367879441171 in Gnumeric 1.12.57.
Availability and version history
- Introduced: Introduced in Gnumeric 1.12.25 release notes.
- Supported in: Gnumeric 1.12.62; no retirement is established.
- Scope: Desktop Gnumeric with its Statistical Functions From R plugin. English comma-separated formulas are shown.
- Checked: October 14, 2026. Tables and the named advanced check use 1.12.57 ssconvert. Current 1.12.62 source was inspected; that release was not executed.
What R.PGUMBEL returns
With z=(x−location)/scale, the lower CDF is EXP(−EXP(−z)). Upper-tail mode gives its complement. Optional log_p selects the natural logarithm of the requested probability. This is the maximum-type Gumbel convention. Versioned function help.
Syntax and arguments
=R.PGUMBEL(x,location,scale[,lower_tail[,log_p]])
x, location and positive scale are required. lower_tail defaults to TRUE and log_p to FALSE. Negative observations and negative locations are valid; a nonpositive scale is not.
Brackets mark optional arguments; do not type them. Current worksheet implementation.
Reproducible examples
These self-contained formulas were recalculated natively. Ordinary-domain results were checked against independent mathematics; source guards explain the boundary cases. Results are rounded to twelve decimal places, or thirteen significant digits for small nonzero values. This is presentation precision, not a universal accuracy guarantee.
| Formula | Gnumeric 1.12.57 result |
|---|---|
=R.PGUMBEL(0,0,1) |
0.367879441171 |
=R.PGUMBEL(1,0,1) |
0.692200627555 |
=R.PGUMBEL(2.5,2,1) |
0.545239211893 |
=R.PGUMBEL(0,0,1,FALSE()) |
0.632120558829 |
=R.PGUMBEL(1,0,1,TRUE(),TRUE()) |
-0.367879441171 |
=R.PGUMBEL(1,0,1,FALSE(),TRUE()) |
-1.178307096421 |
=R.PGUMBEL(-2,-1,.5) |
0.000617978989 |
=R.PGUMBEL(2,2,3) |
0.367879441171 |
Practical worksheet
At the location itself, the lower probability is EXP(−1), approximately 0.367879441171, rather than 0.5. With location 0 and scale 1, x=1 has lower probability approximately 0.692200627555.
For location 2, scale 1 and cutoff 2.5, the CDF is approximately 0.545239211893. The upper and lower calls at that cutoff add to one in the saved advanced check. Location is therefore not a median parameter.
Advanced interpretation and checks
The lower log-tail is exactly −EXP(−z) mathematically. At x=1 under the standard parameters, it is approximately −0.367879441171. The upper log-tail is different: LN(1−EXP(−EXP(−z))).
For a far-left observation x=−100, the tested ordinary lower probability became zero while its log remained finite. For far-right x=1000, however, the internal EXP(−z) underflow caused the upper log-tail failure in the warning. Requesting a log result does not automatically avoid every intermediate underflow.
The saved advanced-check result for =R.PGUMBEL(2.5,2,1)+R.PGUMBEL(2.5,2,1,FALSE()) was 1.
Boundaries and troubleshooting
Zero or negative scale gives #NUM!. The tested extreme ordinary probabilities include rounded zero and one; these are finite-machine outcomes for a distribution whose finite-cutoff probabilities are mathematically between them. The right-tail failure is separately retained rather than described as an invalid mathematical probability input.
Numeric text is parsed, nonnumeric text gives #VALUE!, and empty F5 becomes x=0. Wrong arity gives #N/A. The current 1.12.62 helper is byte-identical to the inspected tested-version helper, but its numerical execution remains unverified.
F5 is genuinely empty, rather than a cell containing an empty string.
| Formula | Gnumeric 1.12.57 result |
|---|---|
=R.PGUMBEL(0,0,0) |
#NUM! |
=R.PGUMBEL(0,0,-1) |
#NUM! |
=R.PGUMBEL(-100,0,1) |
0 |
=R.PGUMBEL(-100,0,1,FALSE()) |
1 |
=R.PGUMBEL(-100,0,1,TRUE(),TRUE()) |
-2.688117141816E+43 |
=R.PGUMBEL(1000,0,1,FALSE()) |
0 |
=R.PGUMBEL(1000,0,1,FALSE(),TRUE()) |
#NUM! |
=R.PGUMBEL("abc",0,1) |
#VALUE! |
=R.PGUMBEL(F5,0,1) |
0.367879441171 |
=R.PGUMBEL(0,0) |
#N/A |
=R.PGUMBEL(0,0,1,0,0,1) |
#N/A |
Official evidence and checking scope
- Current worksheet callback
- Tested-version worksheet callback
- Current pgumbel calculation
- Tested-version pgumbel calculation
- Nonfinite worksheet-value handling
- Statistical Functions From R plugin declaration
The relevant complete worksheet callback and main calculation helper are unchanged between the inspected versions. That does not establish later-version numerical execution. Both stale and uncached SUM controls recalculated correctly. Native XML preserves the R.* identities. The secondary ODF export keeps these calls in the ORG.GNUMERIC namespace; it does not establish execution in another product. GUI entry, other locales and other spreadsheet products were not tested.