The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> Circuit Simulators >> number of runs in Monte carlo simulation
https://designers-guide.org/forum/YaBB.pl?num=1234262422

Message started by filipe on Feb 10^th, 2009, 2:40am

Title: number of runs in Monte carlo simulation
Post by filipe on Feb 10^th, 2009, 2:40am

Hi,

How to estimate the minimum number of runs in a Monte carlo simulation???
I heard that this estimation can be calculated based on the central limit theorem.

Thanks
Filipe

Title: Re: number of runs in Monte carlo simulation
Post by boe on Feb 10^th, 2009, 3:59am

Hi Filipe,
sigma/sqrt(N) is a measure for the accuracy of the average because averaging N normally distributed random variables reduces the standard deviation by a factor of sqrt(N). [Edit: derived from Central limit theorem]
BOE

Title: Re: number of runs in Monte carlo simulation
Post by vivkr on Feb 12^th, 2009, 12:16am

Maybe a more concrete number for you. Typically, we are trying to estimate the moments of the distribution for a given parameter (mean and standard deviation) within a certain confidence level.

As a benchmark, 100 runs of Monte Carlo will typically give +/-2% accuracy in estimation of mean, and +/- 14% accuracy in estimation of the standard deviation. Doubling the number of runs will improve the accuracy by sqrt(2).

Best regards,

Vivek

Title: Re: number of runs in Monte carlo simulation
Post by Andrew Beckett on Feb 12^th, 2009, 1:20pm

Note also that spectre supports "Latin Hypercube Sampling" these days which tends to converge the standard deviation with fewer samples http://en.wikipedia.org/wiki/Latin_hypercube.

Regards,

Andrew.

Title: Re: number of runs in Monte carlo simulation
Post by filipe on Feb 16^th, 2009, 10:31am

Vivkr,

This is exactly my question. How can I calculate the accuracy of the mean and standard deviation? When can I find an explanation about it?

Thanks a lot
Filipe

vivkr wrote on Feb 12^th, 2009, 12:16am:

Title: Re: number of runs in Monte carlo simulation
Post by packiaraj on Feb 27^th, 2009, 6:09am

Filipe,

Guess you must have seen this link by now: ::)

http://www.financial-risk-manager.com/risks/market/mc_errors.html

The expression for error-in-"Mean" that you get from MC sims. BOE has indicated it almost correctly, though it is for only 1Sigma (~68% confidence-level in Normal Distribution).

Think we have to add this error-in-"Mean" to actual "Mean" when we say this is my Bandgap-output or Oscillator-frequency accuracy.

Title: Re: number of runs in Monte carlo simulation
Post by analogue_guy on Mar 2^nd, 2009, 12:43pm

filipe wrote on Feb 16^th, 2009, 10:31am:

Vivkr,

This is exactly my question. How can I calculate the accuracy of the mean and standard deviation? When can I find an explanation about it?

Thanks a lot
Filipe

After running the Monte Carlos and collecting the results you can calculate confidence interval for mean,sigma.

One is based on Student-t deviation, the other is based on CHI-SQUARE deviation. You can find more details in every basic statistics books.

If you would like, I could look up the equations.

Cheers

Title: Re: number of runs in Monte carlo simulation
Post by jiesteve on Mar 19^th, 2009, 4:43pm

analogue_guy,

Could you please send me or post the equations and/or a reference that has this information? I'm also interested.

Thanks!