# Is the probability zero?

1. Jun 17, 2008

### aquaregia

Something that I have always wondered: say you know that a robot will push a button during a 2 minute period after a timer has been started, and you know that the robot picks a time to press it at complete random.

Is the probability that the button will be pressed exactly 1 minute after the timer is started zero? I would say this because probability is defined as (# of specified events/# of possible events). Assuming time and movement are continuous, you would have infinite possible events, and 1 specified event (the timer is exactly 1.000...), and 1/infinity=0. But the robot has to press it at some time so say it presses it at exactly the sqrt2 minutes. Before it happened, the probability that the robot would press the button at sqrt2 minutes was 0 but then the event happened. How is that possible?

2. Jun 17, 2008

### matt grime

This is a question about continuous random variables (and measures). Assuming that the period of 2 minutes is infinitely divisble, then the probability that something happens at precisely x seconds (of the 120) is zero under the uniform measure. That's because we only can talk about an event happening within some subset of the period - the measure of a point is zero.

But don't forgot that there is a difference between reality and a mathematical model of reality.

3. Jun 17, 2008

### aquaregia

But there is some exact single instant when the button is finally pressed (at least in a non-quantum mathematical universe), right? Or say we were talking about a perfect mathematical sphere getting dropped onto a perfect plane. There would be some exact instant when the sphere first touched the plane and that time could be any real number. Or a better example might be considering a random length perfect stick which could be any real number between 1 and 2. How could the probability that its length be 1.5 be zero, when it actually COULD be 1.5?

4. Jun 17, 2008

### shamrock5585

because nothing can be exactly 1.5 based on our measuring system... is it 1.50000000000000000000001 or is it 1.50000000000000000000000000000000000000000000000000001

it can never be perfect unless there is 1.5 and an infinite number of zeros and infinity is not a real number. This is why humans can never create a "perfect circle" because pi is a never ending decimal

5. Jun 17, 2008

### maverick_starstrider

yes but if time is a real number then between any two instants in time (no matter how close together they are) there are an infinite number of other possible times (i.e. between 1.000000000 and 1.000000001 there is also1.0000000005, 1.0000000001, 1.00000000055, etc.) therefore the probability of 1 exact instance in an infinite number of possiblities is $$\frac{1}{\infty}$$ which is zero.

6. Jun 17, 2008

### matt grime

there is, in reality, but that has nothing to do with the fact that you're attempting to model things with probaility theory, and if you choose a continuous r.v. on the interval [0,2] with the uniform distribution, then the probability that it happens at 1, or any other singleton set is zero.

The rest of your post confirms that you don't understand that in continuous r.v.s the probability zero does not mean that it cannot happen. Sorry about the number of negatives in that statement - there is a very significant difference between continuous r.v.s and discrete r.v.s.

7. Jun 17, 2008

### shamrock5585

maverick i see what your saying but its not 1 divided by infinity... what i was explaining is that there is no EXACT instance... to have an EXACT instance you would need an infinite amount of zeros after that number... anything that occurs in time-space has a duration! a true impulse is 1/0 which does not exist in reality!

Last edited: Jun 17, 2008
8. Jun 17, 2008

### CRGreathouse

It is a misunderstanding that events of probability 0 are impossible.

9. Jun 17, 2008

### maverick_starstrider

It need not. Definitions are funny like that. In this case I can say my 'event' is defined as the point in space-time where the variable t=1 (exactly 1) and there you go. It's the time which is the complement of the set (-infinity,1),(1,infinity).

10. Jun 17, 2008

### matt grime

Thanks CRGreathouse, for stating my point with a minimal number of negatives. Wish I'd've stated it like that.

11. Jun 17, 2008

### shamrock5585

yes but does anything in reality actually occur at an actual instance or does it have a duration, even if it is in nanoseconds.

12. Jun 17, 2008

### maverick_starstrider

I dunno. Ultimately a space-time 'event' is all about abstract notion. Therefore, unless you can say in some irrefutable way the time... say... it takes a wavefunction to collapse then an event can often be nothing but an abstract notion and that's fine.

13. Jun 17, 2008

### gel

Mathematically you can suppose that the event occurs at a precise time in the interval [0,2]. The answer to the question is down to http://en.wikipedia.org/wiki/Cantor's_first_uncountability_proof" of the real numbers .
Given a sequence of events A1, A2, ... each with probability 0, then their union also has probability 0.

However, this is not true for an uncountable collection of events.
For such an uncountable collection each with probability 0, you can't infer that their union has probability 0.

This is down to http://en.wikipedia.org/wiki/Sigma_additivity" of probability measures.

Last edited by a moderator: Apr 23, 2017
14. Jun 18, 2008

### aquaregia

15. Jun 22, 2008

### DeaconJohn

aquaregia,

Good point. Good question. Well put! For myself, I am primarily an applied mathematician. Have been since an early age. So, your question strikes home to me.

Yes, of course, as you explain, there are very good reasons for declaring the probability is zero that the even will occur between 732 and 733 seconds after the start time (assuming that all one second intervals are equally likely). My guess is that there is probably a mathematical theory lying around somewhere that develops a theory of probability that covers this point. Probably even "non-standard analysis" addresses it, but I'll bet there is something simpler, and more relevant to real life, than non-standard analysis.

I'm not sure, though. There is a scale free approach to Baysian statistics that claims to address such questions. The conclusion there, however, is that not all one second intervals are equally likely. Roughly, they claim that a one second interval twice as far in the future as another interval is half as likely as the closer one. I don't like that aproach, the classic "tram problem" shomws that it has technical flaws. Probably there are other approaches. Probably no known approach is perfect.

Deacon John

16. Jun 22, 2008

### D H

Staff Emeritus
The probability distribution function for a continuous random variable is used to measure the probability of an event occurring over some range of values (a range with a non-zero measure, to be precise). The probability is defined as

$$P(X\in(x_0,x_1)) = \int_{x_0}^{x_1} p(x)\, dx$$

If you use this definition to compute the probability of hitting an exact point you will always get zero because $\int_a^a f(x)\, dx = 0[/tex] for any function that is integrable over some neighborhood of a. As Matt and GR mentioned above, this does not mean that the robot cannot punch a button. A zero probability does not necessarily mean an impossible event. It might just mean that you erroneously used the pdf to calculate the probability of an event occurring over a space of measure zero (which is exactly what you are doing). That said, there are ways to use the probability distribution function to predict when the robot will punch the button. The pdf lets you compute statistical measures such as the mean, median, variance, etc. The probability distribution no longer applies once the robot has punched the button. He punched the button at exactly 1.414... minutes. There is no uncertainty. The probability he did punch the button at 1.414... minutes is exactly one, and the probability he punched it at any other time is identically zero. One way to visualize what happens when the robot punches the button at 1.414... minutes is to use the approach many physicists take. Envision the pdf as collapsing to the Dirac delta distribution [itex]\delta(x-\surd 2)$ the instant the robot punches the button at 1.414... minutes. Physicists deal with a similar conundrum in quantum mechanics. The outcome of quantum mechanical problems is probabilistic, dictated by the wave function for the process at hand. Once a measurement is made the only uncertainty left is that inherent in the measurement process. One interpretation of quantum mechanics is that the wave function collapses to that measurement uncertainty.

17. Jun 22, 2008

### DeaconJohn

Thanks D H,

I had mis-understood the question. It was more of a "beginner's question" than I thought. No, that's no quite right. It was a question with a simpler explanation than I thought. Unfortunately, it reminded me of a related question (stated in my post) that does not have such an easy answer, as far as I know.

Deacon John

18. Jun 23, 2008

### krikker

1/infinity isn't zero. It's infinately small, which for most intents and purposes is zero.

19. Jun 23, 2008

### shamrock5585

krikker dont argue that... in a math class you are correct... but in this forum some people will argue stuff like that to the death haha i know from experience.

20. Jun 23, 2008

### D H

Staff Emeritus
Whether you are in a math class or a physics class, saying "1/infinity isn't zero. It's infinitely small..." is not correct. In a math class, "1/infinity" is not defined, end of story. In a physics class, 1/infinity is short for $\lim_{x\to\infty} \frac 1 x$, and this is zero.