LOGINV Function (Gnumeric)

• Introduced: First release not established from the evidence cited here.
• Supported in: Gnumeric 1.12.62, confirmed by its function registration. No retirement is established.
• Scope: Desktop Gnumeric with the Statistics Functions plugin. Tables use English comma-separated formulas.
• Checked: October 14, 2026. Tables were recalculated in 1.12.57 and finite reference values independently checked. Version 1.12.62 was source-checked, not executed.

LOGINV returns a left-tail lognormal quantile. If LN(X) is normally distributed with mean μ and standard deviation σ, the quantile is exp(μ+σ×Φ⁻¹(p)). The parameters describe the logarithms of the measurements. In particular, exp(μ) is the median, not generally the arithmetic mean. The arithmetic mean is exp(μ+σ²/2). Official function reference.

=LOGINV(p,mean,stddev)

All arguments shown are required. The version-pinned help and evaluator establish the Gnumeric contract.

The table records the tested release’s outputs, with finite mathematical references rounded to twelve decimal places. Keep full precision when using a result in a later calculation.

For a positive-duration model with log-mean 1 and log-SD 0.5, enter 0.9, 1 and 0.5 in A2:C2. =LOGINV(A2,B2,C2) gives a 90th-percentile duration of approximately 5.159170355623. The output has the measurement units used to define the model. A probability of 90% means 0.9, while the number 90 is invalid. A percentile is a model threshold, not a guarantee about a particular observation.

If the arithmetic mean is 10 and arithmetic SD is 5, first calculate =SQRT(LN(1+(5/10)^2)) for σ and =LN(10)-LN(1+(5/10)^2)/2 for μ. The results are approximately 0.472380727077 and 2.191013317337. Supplying their unrounded values to LOGINV at p=0.95 gives approximately 19.453178380092.

For an ordinary inverse check with μ=1 and σ=0.5, feed LOGINV’s unrounded output to LOGNORMDIST with the same parameters. Be careful with negative μ: LOGINV accepts it, but the Gnumeric LOGNORMDIST wrapper rejects it, so that inverse check is not available across the entire mathematical lognormal parameter domain.

All three arguments are required; do not import another product’s omitted-mean or omitted-SD defaults. The tested p=0 result is zero, while p=1 produces #NUM! rather than an infinite displayed percentile. The wrapper admits both endpoints, but its unbounded upper result is rejected when converted to a worksheet value. Negative log-means are allowed here; the example with μ=−1 has median exp(−1). Nonfinite numerical results are converted to worksheet errors.

These additional outcomes were observed in the same native recalculation. F5 was genuinely empty; it is not an omitted argument. Missing inputs still need deliberate validation, even if they are converted to a numeric zero.

The worked calculations do not certify every locale, imported text value, extreme parameter or later binary version. Shared numerical helpers can change even when a function’s wrapper remains the same.

• Gnumeric function reference
• Gnumeric 1.12.62 evaluator

The reference and boundary tables use recorded 1.12.57 command-line recalculation. Further model examples were independently calculated from their stated inputs; they are not additional native test claims.