The answer depends on several factors. For numerical solutions that are relatively new, difficult, and are related to engineering problems, differences of up to 30% are sometimes still acceptable....

This depends upon the experiment accuracy and the range of numerical model simulate the research problem. However, belo 10% is good.

The best values between 5 to 10%

Your question is a bit ill-posed, since you have not defined in which variable the error occurs. So the correct answer to this question would be: 'it depends'. More context is needed to give a definitive answer.

I agree with Leander's answer. Also, I recommend to read the ASME V&V 20 Code. I have used this code for many verification and validation process of the simulation tools.

The answer depends on several factors. For numerical solutions that are relatively new, difficult, and are related to engineering problems, differences of up to 30% are sometimes still acceptable....

I think to the good numerical results, the error must be less or equal 5% , we

one can see this in several fields such as physics (when we use the finite element method, MOM, FDTD, ...) , numerical analysis or statistics ( confidence interval) ...

It is generally believed that the error is less than 10%.

I hope by validation you mean to compare your numerical results with experimental data. In such case you should rely on experiments outputs provided these data have already been judged. Error within 5 percent is acceptable.

Dear Mahmoud Souby

As far as I'm concerned, we cannot define a specific range for sure, since each case has its own features. The point is properly to set up CFD not only as an operator but using the relevant foundation in the right fashion, then whatever the result comes, it would be appreciable scientifically. There is a pitfall here, for some people are laying emphasis on an identical percentage concerning the standard difference between CFD and experimental test which may lead to manipulation or even unintentionally bias in the CFD results. but if you ask me we should stay away from it. There is one thing here that might be helpful besides the foundation which is the previous literature. They could give us an insight into the case, however, we had better only stick to the foundation and right set up of our task.

Regards

There is no definite range to be recommended. Usually there should be either under or over prediction of numerical solution if compared to experimental or other's results. Based on many parameters and physical understanding of the problem, the researcher should validate and judge his/her solution. Most importantly, numerical results may seem reasonable qualitatively, however, it is not acceptable quantitatively. So be very cautious when carrying computational analysis.

A good way to view numerical predictions is in an engineering context. If you ask me whether the velocity in a room predicted to be 1m/s is accurate I do not know ,it may be 1.3 ,but if you ask me what the result is going to be given a design change ,for example put a short wall hanging down from the ceiling I will guarantee that the relative change of the velocity ,say 10% will be accurate!!

Totally agree with the replies "there is no definite range to recommend". this is generally depends on the associated uncertainties in experimental result itself, Experimental results as numerical results implies simplifications and assumptions and not completely reflect the real situations. So, you have to take the uncertainty budget into your considerations when judging the agreement and hence the acceptable difference between simulation and experimental findings.