minimum particle radius limit?

Junta_20361

hi expert,

I'm trying to use EDEM for a powder compaction simulation.

May I ask if there is any limitation on the minimum radius of the particle in EDEM?

Because when I tried to enter the radius as 1 micrometer, it shows the error as below

I couldn't find any information from EDEM help.

Thank you!

Joshua García

Hello, must be other problem becouse I had being use micrometers with out problem.

jpmorr

It's not an error as such, EDEM whould still run in some cases, but it is a warning that you should pay attention to.

Your mass, volume and moments are incredibly small numbers (10^-14-27) and you will not have much accuracy at all. You will get lots of floating point inaccuracy at this level of precision, which is what I think EDEM is warning about. You should change your units to more suitbale ones for particles of this size, such as um, mg and mm^3 or um^3 and gcm^2. This should remove the warning and ensure that you have results that can be deemed reliable.

 

 

RWood

Hi,

Although the text changes to red, there aren't any errors with modelling like this. I'm not sure of the historical reasons why the colour change was implemented for the values it is, but you should be able to model particles of this size without issue. As Joshua comments, you can go down to micrometers ok, though nanometers might introduce rounding issues somewhere. EDEM should still run though. 

With EDEM 2021.2 (to be released imminently) you'll get the choice of a (fully) single, double or hybrid precision solver. Even a single precision solver can handle values up to +/-1e+/-38 , which you are not hitting here, so you should be ok.

Cheers,

Richard

Junta_20361

hi guys, 

Thank you for your detail answers.

It very clear and helpful.

jpmorr

Richard Wood_20774 said:

Hi,

Although the text changes to red, there aren't any errors with modelling like this. I'm not sure of the historical reasons why the colour change was implemented for the values it is, but you should be able to model particles of this size without issue. As Joshua comments, you can go down to micrometers ok, though nanometers might introduce rounding issues somewhere. EDEM should still run though. 

With EDEM 2021.2 (to be released imminently) you'll get the choice of a (fully) single, double or hybrid precision solver. Even a single precision solver can handle values up to +/-1e+/-38 , which you are not hitting here, so you should be ok.

Cheers,

Richard

While the maximum value that can be stored in a float is of the order of e^38 and for doubles the max value is of the order of e^308, the precision is significantly lower.

 

Single precision floats are accurate to approximately 8 decimal places and doubles are accurate to about 15 decimal places (this varies slightly). Therefore, you should always try to ensure that you are working well within the precision of the machine, hence the EDEM red text warning. Changing the units will change the number of decimals required and improve the precision of your calculation by not storing all the unnecessary zeros.

 

https://en.wikipedia.org/wiki/Double-precision_floating-point_format

RWood

I had a chat with our dev team and wanted to mention a few things, for clarity.

With any recent EDEM build both CPU and GPU solvers are fully double precision. It’s only with EDEM 2021.2 that we’re bringing in this option for single/hybrid/double for our CUDA solver. The CPU solver will remain fully double precision.

Changing the units doesn’t have any impact on the solver, as everything is converted to SI units. The unit choice is purely for visualisation, so this won’t affect the number of zeroes.

With the values seen in the image we’re not even getting close to double precision issues. As you mention JP, there’ll be ~15 -16 significant digits in the mantissa but with values of order e+/-308, any realistic values aren’t going to reach this.  

Which then brings us to the red text. This isn’t related to any kind of precision limit. I’d initially assumed it was as well, but the change in colour is prompted by having a certain value reached. It seems a bit arbitrary but I believe is just what is considered ‘normal’ values.

Long story short, the values here are nothing to be concerned about, unless you choose to model using the single precision CUDA solver in EDEM 2021.2, at which point you might come close to issues.

Richard