RANDEXPPOW Function (Gnumeric)
Generate symmetric exponential-power variates in Gnumeric with RANDEXPPOW, scale/exponent roles, normal and Laplace cases and version caveats.
Verification status: Gnumeric 1.12.57 ssconvert bounded RANDEXPPOW fixtures; pinned/retained 1.12.62 source separate; no GUI. How formula examples are checked
RANDEXPPOW generates a random value from a symmetric exponential-power distribution centered at zero. Its exponent changes the shape of the tails and central peak. It is not the ordinary positive-only exponential distribution, despite the similar name.
Syntax and parameter domain
=RANDEXPPOW(a,b)
Both arguments are required, with no defaults. a is a scale parameter and b is a positive exponent. For a nondegenerate distribution use a > 0 and b > 0. The inspected generator accepts a = 0 as a degenerate zero result; negative a or nonpositive b is invalid.
For positive parameters, the density is proportional to EXP(-ABS(x/a)^b), normalized by 2*a*GAMMA(1+1/b). The distribution is symmetric and can produce negative as well as positive values. Its scale a is not generally its standard deviation.
The function is volatile: recalculation can generate a new value. There is no seed parameter or promised repeatable random sequence in this syntax.
Worked example: a familiar normal special case
Enter =RANDEXPPOW(2,2) in A1. The expected output is a random numeric value, with no fixed individual value. For b = 2, the distribution is normal with mean zero and standard deviation a/√2. Here the theoretical standard deviation is √2, approximately 1.414213562373, and the theoretical variance is 2.
For a boundary check, =RANDEXPPOW(0,2) returned #NUM! in native Gnumeric 1.12.57, whereas the retained 1.12.62 source explicitly returns 0 for zero scale. This is a version-specific difference, not a verified 1.12.62 execution. Use strictly positive scale for a nondegenerate example portable across these inspected implementations.
=RANDEXPPOW(-1,2) and =RANDEXPPOW(1,0) produce #NUM! in the bounded native fixtures. The parameter errors must be corrected; replacing them with zero would silently change the modeled distribution.
Advanced use: compare shape at a common variance
For b = 1, the distribution becomes a zero-centered Laplace distribution with variance 2a². Therefore =RANDEXPPOW(1,1) and =RANDEXPPOW(2,2) both have theoretical variance 2, even though their shapes differ. The first has Laplace tails, the second normal tails.
This is a useful comparison when the goal is to change distributional shape without also changing the theoretical variance. Equal parameter a would not accomplish that for these two exponents. For general b, variance is a²*GAMMA(3/b)/GAMMA(1/b); this identity also reproduces both special cases independently.
A sample mean or sample standard deviation will fluctuate. A small sample will not equal the theoretical moments exactly, and a single draw cannot validate a distribution. If you need a fixed realization for a report, preserve the generated numeric values rather than leaving all cells as volatile formulas.
Errors and practical cautions
- Both arguments are required; b is not an optional rate or a probability.
- Use b > 0 and a > 0 for these ordinary examples; negative scale and zero exponent were rejected in the native tests. Zero scale also failed in 1.12.57, although the inspected 1.12.62 source handles it as zero.
- Do not expect all values to be positive. This is a symmetric distribution over the real line.
- a is a distribution scale, not generally the standard deviation. Use the moment formula when calibrating a spread.
- Very small exponents can produce extreme tails and numerical challenges. The bounded native calls here used the well-behaved b = 1 and b = 2 branches, not extreme parameters.
- Random numeric type checks do not certify tail accuracy, independence or cryptographic suitability.
History and related functions
Official sources
Native verification scope
On 17 October 2026, Gnumeric 1.12.57 ssconvert recalculated 5 bounded RANDEXPPOW fixture records in C.UTF-8. The saved input workbook, exact formula cells, expected specification and output workbook were cross-checked against the native result records. Ordinary random calls were tested only for numeric results; deterministic zero/error cases were checked separately. This is not statistical certification or a fixed-draw promise. This is native engine evidence, not GUI or screenshot evidence. Inspected 1.12.62 source is separate from the 1.12.57 executable. Composed helper formulas and distributional identities are mathematical checks unless separately recorded. Other versions and untested boundaries remain outside this scope.
Cross-program scope
These are Gnumeric function names and argument conventions. Similar option or random-distribution functions in other spreadsheets may use different defaults, scaling or parameterizations. Exporting a workbook does not establish that another program will recalculate these formulas; verify support and results in the destination program before relying on them.