Dear Piotr Oskar

Can you give some illustration to show me, what type of result are you looking for. I have some familiarity with EPA PMF 5 but not with PCA; so that i may know if i can be of providing useful advice..

Thanks

Wajih.

Dear Piotr Oskar

First you check mass closure of chemical species w.r.t to PM mass and pretreat your data by MDL, BDL and Missing value as per suggested in Kim and Hopke 2004.

If you have any doubt you can discuss freely. There is better step to follow for input file to PMF run. 

Thanks.

Jai

Sorry for late answer. Back home yesterday.

Problem seems to be simply, so don’t understand why didn’t get answer from EPA: In PCA have I matrix notation: (S-LI)w=0 or Sw=Lw (S-cov. Matrix, L-eigenvalue, w-eigenvector); need to solve pure w vector. In Statistica or other soft get rjl as reults, instead of wjl. Weights. No problem.

In PMF have weights before rotation. Where can I find weights matrix in results? (“A guide to positive matrix factorization” by Philip K. Hopke p.6)

That’s the point. Maybe I didn’t understand something.

I looked at "A Guide To Positive Matrix Factorisation" by Philip K. Hopke. The 'weights' mentioned here are called 'uncertainties'. These can be determined analytically by the laboratory, or with equations from the US EPA:

for concentrations  detection limit (DL): uncertainty = (DL^2 + error fraction^2)^0.5

Factor loadings are known as the source profiles (Huston et al. 2012 - source Source apportionment of heavy metals and ionic contaminants in rainwater tanks in a subtropical urban area in Australia), and may be found as a bar chart underneath the Base Model tab. I am not sure how to extract the original matrix - possibly it is in the output file.

Factor scores are known as source contributions (Huston et al. 2012 - source Source apportionment of heavy metals and ionic contaminants in rainwater tanks in a subtropical urban area in Australia), and may be found in the same tab as above.