Using Cococcioni linear-response U in Dudarev DFT+U (LDAUTYPE = 2)

[tasdiquearman]

Can the Hubbard U obtained from the Cococcioni linear-response method (LDAUTYPE = 3) be directly used as Ueff in the Dudarev DFT+U approach (LDAUTYPE = 2), i.e., by setting LDAUU = ULR and LDAUJ = 0?

[jonathan_lahnsteiner2]

Dear tasdiquearman,

Yes, the Hubbard U obtained from the Cococcioni linear-response method (LDAUTYPE = 3) can be directly used as Ueff in the Dudarev DFT+U approach (LDAUTYPE = 2). Specifically, you set LDAUTYPE = 2, LDAUU = ULR (the value obtained from linear response), and LDAUJ = 0.

*) The Cococcioni method computes only the averaged Hubbard U, i.e., U_eff = U − J .
*) The Dudarev approach requires U_eff as the input, not independent U and J values.
*) With LDAUTYPE=2, only U_eff is used, so you provide the linear response value as LDAUU and set LDAUJ = 0.

Please also take a look at this tutorial which might be of help to you:
https://www.vasp.at/wiki/Calculate_U_for_LSDA%2BU

All the best Jonathan

[tasdiquearman]

Using this method, I obtained an effective Hubbard parameter 𝑈eff = 6.11 eV for Fe in a polyanionic framework (oxide and pyrophosphate), which is noticeably higher than the commonly adopted U values for Fe. How can I assess whether this value is physically meaningful and reliable for my system?

[jonathan_lahnsteiner2]

Dear tasdiquearman,

It is not possible to tell which value of U is reasonable with the information you supply. You could use the computed U value and predict some physical quantity. This property you can compare to experiments or to other computational studies. Based on this it is possible to decide if your U value makes sense.

All the Best Jonathan