how to generate true random number

Hi everyone: I am trying to generate true random number by MATLAB. I tried the intrinsic functions, 'randn'. But finally, I found it is not true random number generators. I generate 10000 random numbers, found the mean of them are not near 0, some cases the mean are 0.04,0.007 ....they are big. And the sum of the numbers is very big, 87, 73 130 ... So I believe that the function 'randn' is not a true random number generator. Does anyone know how to generate true random number by MATLAB?

[Stephen23]

" I am trying to generate true random number by MATLAB"

This is a contradiction in terms: all computers (that MATLAB currently runs on) are deterministic.

Some ways to get "true random numbers":

• Use radioactive decay
• Use a service like https://www.random.org/, which apparently uses atmospheric noise.
• Use diode breakdown noise.
• Buy a book of random numbers.

Or simply accept that the algorithms used by MATLAB are as random as you will need for any modelling that you are doing.

[Walter Roberson]

http://mentalfloss.com/article/81946/7-sources-randomness

Answer 1

[David Goodmanson]

Hello, hu

yes, Matlab random numbers are not truly random, but they are a very good simulation. The results you are seeing would happen even for a set of truly random numbers. Suppose you take N samples from a given distribution with mean m and standard deviation s. By the central limit theorem, for large N those samples have a normal distribution with mean m and standard deviation s/sqrt(n). So you you should expect some variations from the official mean, even for perfectly random numbers. And those variations should be on the order of 1/sqrt(1e4) = .01.

[hu wong]

Hi, David: Thank you for your replying. I got to know even true random number generators can not generate exactly a group of numbers with mean close to 0. I am now trying to use a random walk model to simulate a zooplankton's movement. So I need a random number generators that can generate a group of numbers with sum is near 0. So it seems that the function 'randn' does not meet the need of the model. Do you have any idea about the random walk model?

[Walter Roberson]

Consider https://www.mathworks.com/matlabcentral/fileexchange/9700-random-vectors-with-fixed-sum

[David Goodmanson]

Hello hu

It depends on what you mean by 'close to 0'. In a random walk model, for N steps you expect that the root mean square distance from the starting point will increase like sqrt(N). If the particle moved with constant velocity then the distance would increase like N. Compared to that, sqrt(N) is close to 0.

sqrt(N) behavior leads to diffusion. If you try to make random numbers with a sum that is artificially smaller than that, then you do not have the standard random walk.

Answer 2

[Walter Roberson]

https://www.mathworks.com/matlabcentral/fileexchange/21353-true-random-integer-generator and https://www.mathworks.com/matlabcentral/fileexchange/27849-randi-org--true-random-integer-numbers

[Walter Roberson]

Through the perl interface:

https://metacpan.org/pod/Crypt::Random

http://perl.mines-albi.fr/perl5.8.5/site_perl/5.8.5/sun4-solaris/Math/TrulyRandom.html

Through the python interface:

https://stackoverflow.com/questions/22891583/can-i-generate-authentic-random-number-with-python

But see also https://wiki.openssl.org/index.php/Random_Numbers

You might also be interested in http://www.instructables.com/id/A-Raspberry-Pi-based-Truly-Random-Number-Generator/

[David]

Is it possible for MATLAB to tap the entropy source on x86 machines?

for intel see: _rdseed16_step/ _rdseed32_step/ _rdseed64_step

If not, it would be nice to add this functionality and have random numbers suitable for cryptography.

As a side note, I suspect that the Sigma-Delta ADC topology was a byproduct of developing similar entropy sources.