log (1-var) operation in C

Question

I am trying to code this algorithm. I am stuck in the part of log((1.0-u)/u))/beta; As I understand, I can not get the result of this in C, as it will always return me with negative value log (returning imaginary value). Tried to print the result of log(1-5) for instance, it gives me with Nan. How can I get the result of

double x = (alpha - log((1.0-u)/u))/beta

then?

Would appreciate for any pointers to solve this problem.

Thank you

What do you expect? 10 to what power (real number) would equal -4? — Richard J. Ross III, Mar 30 '12 at 17:30
@OliCharlesworth correct, but remember that log doesn't return a complex number. You'd need a different log function. — Richard J. Ross III, Mar 30 '12 at 17:32
Just kidding ... you can use the [clog() function](http://port70.net/~nsz/c/c99/n1256.html#7.3.7.2). — pmg, Mar 30 '12 at 17:37

score 3 · Accepted Answer · answered Mar 30 '12 at 17:31

3

In that algorithm, u should be uniform random on [0,1], so the quantity (1-u)/u is never negative.

answered Mar 30 '12 at 17:31

Stephen Canon

100,816
18
175
263

Oliver Charlesworth · Answer 2 · 2012-03-30T17:37:27.757

2

Don't pass in a value of u outside the range (0,1) (this is mentioned in one of the comments in that article). Note that ( and ) denote open (i.e. exclusive) bounds.

edited Mar 30 '12 at 17:37

answered Mar 30 '12 at 17:29

Oliver Charlesworth

260,367
30
546
667

1

This is correct; if you read down in the Poisson algorithm you link, it states that the `random()` function listed in the pseudocode returns a number in the range (0,1). However, if you do need to handle complex numbers, you should look at C99 complex number support. See this article for more information: http://stackoverflow.com/questions/6418807/how-to-work-with-complex-numbers-in-c – Ethan Brown Mar 30 '12 at 17:37

score 1 · Answer 3 · answered Mar 30 '12 at 17:37

1

As stated you need the range of u to be (0,1) (if u is 0 or 1 you are in trouble). You are probably using rand() in which case you will want

double u = (rand() + 1) / (double)(RAND_MAX + 2);
double x = (alpha - log((1.0-u)/u))/beta

answered Mar 30 '12 at 17:37

RunHolt

1,832
2
19
26

score 0 · Answer 4 · answered Mar 30 '12 at 17:31

0

I guess u lays between 0 and 1, thus being a normalized positive random.

answered Mar 30 '12 at 17:31

iehrlich

3,552
4
31
42

score 0 · Answer 5 · answered Mar 30 '12 at 17:41

log( (1-u)/u ) = log(1-u)-log(u) and thus u needs to be 0<u<1. The graph looks like this

log(1-u)-log(u)

In fact you see the anti-symmetry by u=0.5 you may need to worry only for values between 0 and 0.5, but not including 0. So for u>0.5 you set u=1-u and return -log(1-u)+log(u).

log (1-var) operation in C

5 Answers5