Contact
12 programs
Explore the site

ERFC in Apple Numbers: Complementary Error Function

Engineering Intermediate Apple Numbers

Understand Apple Numbers ERFC, Gaussian tail calculations, scientific notation, examples and troubleshooting.

Verification status: Official documentation referenced; example outcomes independently reasoned or arithmetically checked where stated. Not executed in Apple Numbers. How formula examples are checked

Quick answer. =ERFC(1) has Apple’s documented output 0.157299207050285. ERFC is the complement to the Gaussian error function and accepts one required numeric argument, not two. Apple’s =ERFC(12) example is approximately 1.3562611692059 × 10⁻⁶⁴, so display formatting can make a nonzero answer look like zero. [A1]

What ERFC calculates

For a real value x, the complementary error function is mathematically

erfc(x) = (2 / SQRT(PI)) × integral from x to infinity of EXP(−t²) dt
        = 1 − erf(x)

This makes ERFC(x) convenient for small Gaussian tail integrals, signal/noise models, diffusion calculations and normal-distribution right tails after the proper change of variables. It is a special mathematical function; the bare value ERFC(x) is not always a probability. For example, the official ERFC(-1) result exceeds 1. [A1, C1]

A crucial numerical difference: calculating 1−ERF(x) when ERF(x) is extremely close to 1 can lose significance through cancellation. Calling ERFC(x) directly is the mathematically conventional way to represent that complement and is often preferable for small tails. This is a general numerical-analysis recommendation; native Numbers precision at particular extreme inputs has not been measured. [C1]

Documented syntax, argument and defaults

=ERFC(lower)
Position Apple argument Type Required Role Default
1 lower Number Yes Lower limit of the complementary Gaussian integral extending to positive infinity None

=ERFC(1,2) is not Apple’s documented syntax. If you need an integral between two finite bounds, investigate ERF(lower,upper) instead. Unlike ERF, ERFC does not have a separately documented optional upper argument. Apple’s help does not describe numeric coercion from arbitrary text or an implied default for a missing lower. [A1, A6]

Build a Numbers table of complementary integrals

Create a table named GaussianTail with B1=Input x, C1=ERFC(x) and D1=ERF(x) + ERFC(x). Set the input values and formulas as follows:

Row B: x C formula Expected C D formula
2 −1 =ERFC(B2) 1.842700792950 =ERF(B2)+C2
3 0 =ERFC(B3) 1 =ERF(B3)+C3
4 1 =ERFC(B4) 0.157299207050 =ERF(B4)+C4
5 2 =ERFC(B5) 0.004677734981 =ERF(B5)+C5
6 3 =ERFC(B6) 0.000022090497 =ERF(B6)+C6

The mathematical expectation in D2:D6 is 1, allowing small floating-point rounding differences. Enter C2 and D2 then fill down to their respective last rows, adjusting cell references. Apple’s published examples include ERFC(-1) and ERFC(1); the other values and D column are independently calculated and not a native Numbers test. [A1, C1]

If entering this on a separate summary table, use the Numbers table-qualified form such as =ERFC(GaussianTail::B4) for the input cell; apply $ anchors where dragging should not move a reference. [A3]

How to see very small results

Apple specifically demonstrates

=ERFC(12)

with expected output 1.3562611692059E-64, and instructs readers to select a Scientific cell data format when the value should be displayed in this notation. If a cell visually says 0, do not immediately conclude the mathematical answer is zero. Formatting, rounding and actual numerical underflow are different phenomena; neither this tutorial nor Apple’s published example establishes the exact underflow threshold of a given Numbers build. [A1]

A useful diagnostic is to compare:

=ERFC(2)
=ERFC(3)

Expected 0.004677734981 and 0.000022090497 respectively. If both display identically as zero because the cell is set to a small number of decimal places, switch to scientific formatting before troubleshooting the formula itself. [C1, A1]

Advanced application 1 — standard-normal upper-tail probabilities

If Z has a standard normal distribution, then mathematically

P(Z >= z) = ERFC(z / SQRT(2)) / 2

For a two-standard-deviation threshold:

=ERFC(2/SQRT(2))/2

Expected 0.022750131948 (~2.275% upper-tail probability). At z=1 the corresponding expression has expected result 0.158655253931 (~15.866%). Neither is the raw output of ERFC(2): the /SQRT(2) argument scaling and outer /2 are essential to turn the error integral into a normal probability. [C1, A1]

To use a table, place a calculated z-score in B2 and enter the right-tail formula in C2:

=ERFC(B2/SQRT(2))/2

For a score x, centre mu and strictly positive population standard deviation sigma, obtain a standardized score first:

=ERFC(((B2-C2)/D2)/SQRT(2))/2

where B2=x, C2=mu and D2=sigma. If D2 is zero or negative, the statistical model is invalid; avoid relying on unspecified spreadsheet handling. [C1]

Advanced application 2 — probability outside symmetric bounds

For a standard-normal Z, a two-sided tail beyond |Z| ≥ 2 is

=ERFC(2/SQRT(2))

Expected 0.045500263896 (~4.55%). Notice that the two-tailed expression lacks the /2 that appears in the one-tailed expression because both symmetric tails are counted. More generally:

=ERFC(ABS(B2)/SQRT(2))

uses a standardized magnitude from B2. This is a normal-model identity and not a license to apply ERFC indiscriminately to skewed, heavy-tailed or otherwise nonnormal datasets. [C1]

Advanced application 3 — short complement identity

Compare the direct complement with an expression through ERF:

=ERFC(1)
=1-ERF(1)

Both have expected result near 0.157299207050285. For large positive x, the second representation may numerically round to zero much earlier than the direct complement. At negative x, ERFC(x) >1 is expected because erf(x)<0; this is not a data-entry mistake. [A1, A6, C1]

Troubleshooting and cautious limitations

Situation Recommended check Evidence
Wrong number of arguments Use exactly one numeric lower argument; don’t supply a second bound to ERFC. Apple syntax. [A1]
Text or blank input Enter a numeric value; Apple’s help does not promise conversion for arbitrary text or blank-cell inputs. Explicit type documentation. [A1]
The result exceeds 1 For negative x, a complementary integral, not a probability, can exceed 1. Apple ERFC(-1) example and math. [A1, C1]
Result appears to be zero Change Numbers’ cell Data Format to Scientific and inspect the exponent. Apple’s ERFC(12) note. [A1]
An upper tail is surprisingly large Check whether the correct normal scaling z/SQRT(2) and probability divisor 2 were applied. Mathematical normal-CDF identity. [C1]
Copy/fill gives the wrong row Confirm whether the intended reference should be anchored like $B$2. Apple cell-reference help. [A3]
Formula seems to have too many/few separators Follow the active Numbers locale; semicolons may replace English comma separators. Apple functions overview. [A2]
Porting Excel/Sheets formulas Compare against Apple’s one-argument ERFC entry rather than assuming another engine’s aliases or edge-case rules. Apple syntax only. [A1]

Apple does not provide a complete documented error-code map, input limit, maximum precision or treatment of nonnumeric coercion in this individual function entry. The tutorial consequently does not invent a specific error token, screenshot, or native acceptance test. [A1]

What is the difference between ERF and ERFC? For one input x, the mathematical identity is ERF(x)+ERFC(x)=1. ERF can also accept a second finite bound, whereas ERFC does not. See Apple’s ERF documentation. [A6]

Why does ERFC(12) not equal 0? Gaussian tails are tiny but mathematically nonzero at finite x. Apple’s displayed answer uses scientific notation with exponent −64. [A1]

Can ERFC values exceed one? Yes, for negative arguments. For instance, Apple explicitly prints ~1.8427 for −1. Do not automatically format every ERFC result as a percentage probability. [A1]

Can I model quality-control defects with ERFC? Only if you have a defensible normal approximation and use the appropriate standardized transformation, not simply ERFC(defects). Check data assumptions independently. [C1]

Does a function appearing in Help 15.4 prove it is new in that release? No. The help version confirms the documented current interface, not the original introduction date. [A1, A4, A5]

Source and independent verification register — 13 October 2026

Native-validation limitation: No Apple Numbers execution or screenshots were performed. Relevant edge cases and locale-specific behavior require testing in the intended application.