#102 Function

Methods for calculating the Gauss error function in C with standard libraries and the GNU Scientific Library.

Notes

In mathematics, the error function is the probability that Y falls in the range [−x, x], where the random value Y is normally distributed with mean 0 and variance 1/2.

Using the C Standard Library

The erf.c example calculates the probability and area under the curve for a couple of sample values. It uses the erf() function from the C standard library, defined in header <math.h>.

The example requires linking with the math library (-lm). The Makefile is setup to compile and link the example, else it can be done by hand:

gcc erf.c -o erf -lm -Wall -O3 -std=gnu11

Installing the GNU Scientific Library

It is probably most common to install the GSL in precompiled form via a package managed. On Mac, I’d do that with brew install gsl - see the gsl homebrew fomula

But for now, I’m going to install from source as follows. NB: 1.15 is a pretty old version now.

wget ftp://ftp.gnu.org/gnu/gsl/gsl-1.15.tar.gz
tar xvzf gsl-1.15.tar.gz
rm gsl-1.15.tar.gz
cd gsl-1.15
./configure
make
sudo make install

All worked fine - installed to /usr/local/lib/libgsl* and /usr/local/include/gsl/, and can now be referenced with the help of pkg-config:

$ pkg-config --libs gsl
-L/usr/local/lib -lgsl -lgslcblas -lm

Using the GNU Scientific Library

The erf_gsl.c example repeats the calculation, but using functions from the GNU Scientific Library, namely:

• gsl_sf_erf
• gsl_cdf_gaussian_P

Running the Examples

$ make
gcc -Wall -O3 -std=gnu11    erf.c  -lm -o erf
gcc -Wall -O3 -std=gnu11    erf_gsl.c  `pkg-config --libs gsl` -o erf_gsl
./erf
The probability that Normal(0, 1) random variable has a value between -0.25 and 0.25 is: 0.276326
The integral of a Normal(0, 1) distribution between -0.25 and 0.25 is: 0.197413
The probability that Normal(0, 1) random variable has a value between -1.96 and 1.96 is: 0.994426
The integral of a Normal(0, 1) distribution between -1.96 and 1.96 is: 0.950004
./erf_gsl
The probability that Normal(0, 1) random variable has a value between -0.25 and 0.25 is: 0.276326
The integral of a Normal(0, 1) distribution between -0.25 and 0.25 is: 0.197413
The probability that Normal(0, 1) random variable has a value between -1.96 and 1.96 is: 0.994426
The integral of a Normal(0, 1) distribution between -1.96 and 1.96 is: 0.950004

Credits and References

• error function - wikipedia
• 21st Century C
• GSL - GNU Scientific Library
• erf() - C standard library
• erf() - GNu libc manual
• gsl homebrew fomula