What's the difference between PBE and B3LYP methods?
PBE
The PBE functional${}^{[1]}$ belongs to the class of generalized gradient approximation (GGA) functionals for the exchangecorrelation energy $E_{\mathrm{xc}}$. Considering that the dependence $E_{\mathrm{xc}}[\rho]$ may be nonlocal, i.e. $E_{\mathrm{xc}}$ may depend on the density $\rho$ at a given point (locality), but also on $\rho$ nearby (nonlocality) the assumption made by the local spin density approximation${}^{[2,3]}$ (LSDA) that $E_{\mathrm{xc}}[\rho]$ is strictly local leaves a lot room for improvement. But still LSDA is a good starting point since it is a rather simple efficient model that gives good results. The GGA functionals improve on LSDA in a way that's pretty much in the spirit of a Taylor series with LSDA as a starting point: When we are at a point, what happens further off depends not only on $\rho$ at that point, but also the gradient of $\rho$ at the point, etc. (much as in a Taylor series: if you want to go further you may need not only the gradient, but also the Laplacian, etc.). So, these functionals are typically constructed by adding gradient corrections to the LSDA functionals. But that is not a trivial thing to do and doing it "naively" leads to wrong results. Therefore, some GGA functionals are parametrized by fitting experimental data. But there are also some GGA functionals that achieve the inclusion of gradient correction without introducing experimentally fitted parameters which makes them valid for a wide range of systems. One example of such a parameterfree GGA functional is the one developped by Perdew, Burke and Ernzerhof (PBE). It is known for its general applicability and gives rather accurate results for a wide range of systems.
Side note: GGA functionals are frequently termed nonlocal functionals in the literature. This is a somewhat misleading and actually sloppy terminology that should be avoided and since my above statements might be misread in that direction I want to make one thing perfectly clear: All GGA functionals are perfectly local in the mathematical sense: the value of the functional at a point $\vec{r}$ depends only on information about the density $\rho(\vec{r})$, its gradient $\nabla\rho(\vec{r})$, and possibly other information at this very point and is absolutely independent of properties of $\rho(\vec{r}^{\, \prime})$ at points $\vec{r}^{\, \prime} \neq \vec{r}$. Calling these functionals 'nonlocal' is only motivated by the fact that these functionals go beyond the 'local' density approximation and of course the observation that knowledge of the gradients is the first step towards accounting for the inhomogeneity of the real density.
B3LYP
The B3LYP approach${}^{[4,5,6]}$ belongs to the hybrid approximations for the exchange–hybrid correlation functional. The approximation is famous, because it gives very good results and, therefore, is extremely popular. The distinguishing feature of such hybrid approximations is that they mix in a certain amount of the exact Hartree–Fock exchange energy into the exchange and correlation obtained from other functionals. There is actually some justification for that: At the lower limit ($\lambda = 0$, where $\lambda$ is the coupling constant) of the so called coupling constant integration the exchangecorrelation hole is equal to the exact exchange hole. This observation led Becke${}^{[7, 8]}$ to conclude that a fraction of exact exchange should be mixed with GGA exchange and correlation. The simplest such hybrid functional is
\begin{align} E^{\mathrm{hyb}}_{\mathrm{xc}} = a E_{\mathrm{x}}^{\mathrm{exact}} + (1 − a) E^{\mathrm{GGA}}_{\mathrm{x}} + E^{\mathrm{GGA}}_{\mathrm{c}} \end{align}
where the constant $a$ can be fitted empirically or estimated theoretically ${}^{[9, 10, 11]}$ as $a \approx \frac{1}{4}$ for molecules. So far so good, but it smells like a witch's brew for the B3LYP exchange–correlation potential $E_{\mathrm{xc}}$: take the exchange–correlation energy from the LSDA method, add a pinch (20%) of the difference between the Hartree–Fock exchange energy $E^{\mathrm{KS}}_{\mathrm{x}}$ (well, in fact, this is Kohn–Sham exchange energy, because the Slater determinant wave function, used to calculate it, is the Kohn–Sham determinant, not the Hartree–Fock one) and the LSDA $E^{\mathrm{LSDA}}_{\mathrm{x}}$. Then, mix well 72% of Becke exchange potential${}^{[12]}$ $E^{\mathrm{B88}}_{\mathrm{x}}$ which includes the 1988 correction, then strew in 81% of the Lee–Yang–Parr correlation potential${}^{[13]}$ $E^{\mathrm{LYP}}_{\mathrm{c}}$. You will like this homeopathic magic potion most if you conclude by putting in 19% of the Vosko–Wilk–Nusair potential$^{[14]}$ $E^{\mathrm{VWN}}_{\mathrm{c}}$:
\begin{align} E^{\mathrm{B3LYP}}_{\mathrm{xc}} = E^{\mathrm{LSDA}}_{\mathrm{xc}} + 0.2 (E^{\mathrm{KS}}_{\mathrm{x}}  E^{\mathrm{LSDA}}_{\mathrm{x}}) + 0.72 E^{\mathrm{B88}}_{\mathrm{x}} + 0.81 E^{\mathrm{LYP}}_{\mathrm{c}} + 0.19 E^{\mathrm{VWN}}_{\mathrm{c}} \end{align}
So, you can see B3LYP contains lots of empirical parameters but that's fine as long as you only want the results and don't want to know why this is working :)
References
${}^{1}$ J. P. Perdew, K. Burke, M. Ernzerhof, "Generalized Gradient Approximation Made Simple", Phys. Rev. Lett. 1996, 77, 3865; J. P. Perdew, K. Burke, M. Ernzerhof, "Erratum to Generalized Gradient Approximation Made Simple", Phys. Rev. Lett. 1997, 78, 1396 (E). (Link)
${}^{2}$ W. Kohn, L. J. Sham, Phys. Rev. 1965, 140, A1133. (Link)
${}^{3}$ U. von Barth, L. Hedin, "A local exchangecorrelation potential for the spin polarized case", J. Phys. C: Solid State Phys. 1972, 5, 1629–1642. (Link)
${}^{4}$ A. D. Becke, "Densityfunctional thermochemistry. III. The role of exact exchange", J. Chem. Phys. 1993, 98, 56485652. (Link)
${}^{5}$ P. J. Stephens, F. J. Devlin, C. F. Chabalowski, M. J. Frisch, "Ab Initio Calculation of Vibrational Absorption and Circular Dichroism Spectra Using Density Functional Force Fields", J. Phys. Chem. 1994, 98, 11623–11627. (Link)
${}^{6}$ K. Kim, K. D. Jordan, "Comparison of Density Functional and MP2 Calculations on the Water Monomer and Dimer", J. Phys. Chem. 1994, 98, 10089–10094. (Link)
${}^{7}$ A. D. Becke, "A new mixing of Hartree–Fock and local density‐functional theories ", J. Chem. Phys., 1993, 98, 1372. (Link)
${}^{8}$ A. D. Becke, "Density‐functional thermochemistry. III. The role of exact exchange", J. Chem. Phys., 1993, 98, 5648. (Link)
${}^{9}$ J. P. Perdew, M. Ernzerhof, and K. Burke, "Rationale for mixing exact exchange with density functional approximations", J. Chem. Phys., 1996 105, 9982. (Link)
${}^{10}$ C. Adamo and V. Barone, "Toward reliable density functional methods without adjustable parameters: The PBE0 model", J. Chem. Phys., 1999 110, 6158. (Link)
${}^{11}$ M. Ernzerhof and G. E. Scuseria, "Assessment of the Perdew–Burke–Ernzerhof exchangecorrelation functional", J. Chem. Phys., 1999, 110, 5029. (Link)
${}^{12}$ A. D. Becke. "Densityfunctional exchangeenergy approximation with correct asymptotic behaviour", Phys. Rev. A 1988, 38, 30983100. (Link)
${}^{13}$ C. Lee, W. Yang, R. G. Parr, "Development of the ColleSalvetti correlationenergy formula into a functional of the electron density", Phys. Rev. B, 1988, 37, 785789. (Link)
$^{14}$ S. Vosko, L. Wilk, M. Nusair, "Accurate spindependent electronliquid correlation energies for local spin density calculations: a critical analysis", Can. J. Phys. 1980, 58, 12001211. (Link)
Related videos on Youtube
Basia
Updated on September 27, 2020Comments

Basia about 3 years
I can't find an answer to that question. I was told that in B3LYP, more variables implemented in the method are empirical, but I can't find anywhere if it's true, and I'm sure it's not the only difference. I'm new to calculations, and it seems so hard and confusing, so please help me.

Aesin over 9 yearsThe Gaussian manual page on how to make it use different functionals is a good source of references on this.

Philipp over 9 years@Basia I updated my answer to say something about locality of PBE and to add references to the seminal papers.


Martin  マーチン over 9 yearsMany people say that pbe is a "nonlocal" functional, but that is not true. It depends also on the gradient, which is a local property of the PES on any given point. I also wished people would not rely on B3LYP so much  it was an improvement back in the day, but there are better and more efficient functionals nowadays.// I also want to state that PBE consists of (at least) two functionals: One for the exchange and one for the correlation, the combination of those two is often denoted as PBE0.

Martin  マーチン over 9 yearsWould you mind if I edit in the complete references?

Basia over 9 yearsI use pbe1pbe functional with 6311+G(2d,p) basis set for all my calculations. I'm generally interested in reactions of perhalogenated compounds. Many people say I should switch to b3lyp because it's better and always gives good results, but after I read your answer I think I'll stick with pbe

Philipp over 9 years@Martin I would be thankful if you took the trouble of editing in the references. I refrained from most of it because I thought it should be rather easy to find. Sorry about the "nonlocal" part: Of course, you are right: PBE is local. But it was late and I wrote the answer before going to sleep so I was a bit sloppy in that point and didn't state explicitly that PBE is not a nonlocal functional. Also, I didn't want to dwell on too many details.

Philipp over 9 years@Martin I added a paragraph making clear that PBE is local. I hope it's clear now. Are you sure about the PBE0 thing? Isn't the name PBE0 usually used as a synonym for the PBE1PBE hybrid functional?

Philipp over 9 years@Basia I didn't want to preach against B3LYP. Actually there is some physical motivation for mixing in around 20 or 25% of HartreeFock exchange into the exchangecorrelation functional. And if B3LYP gives you better results than PBE I see little reason not to use it. But as Martin stated there are better functionals than B3LYP out there although he might be better suited than I am to recommend one to you. I will only mention that if hybrid functionals are better suited for the systems you calculate you could try the PBE1PBE hybrid functional (also called PBE0) which has fewer parameters.

Philipp over 9 years@Martin I added the references.

Aesin over 9 years@Philipp: Agreed, whenever I've seen the name PBE0 it's been for the modified hybrid version of PBE that includes weighting factors and a fraction of HF exchange, as opposed to pure (GGA) PBE which doesn't use any HF exchange.

Martin  マーチン over 9 years@Aesin I was confused myself  I am very sorry, of course PBE0 is the modified Hybrid functional. @ Philipp I like your answer so very much and while reading it the first time I thought it was not sloppy  I just wanted to point out a very common (very unfortunate) misconception. (And extra kudos for the witch's brew comment.)

Martin  マーチン over 9 years@Basia I am not a big fan of b3lyp and I tend towards much more robust (BP86) or much more sophisticated functionals (PBE, Minnesota, ...). As with all calculations they have to account for a certain standard and using one df is usually not enough. The problem is that a functional that is good for describing one bonding situation might be terrible for describing another. There is unfortunately no best functional. Stick with pbe1pbe when you started with it, and confirm results with other functionals or ab initio. I highly recommend switching to a more sophisticated basis set.

Philipp over 9 years@Martin Nice to hear that you like my answer, thanks. And thanks for pointing out the misconception of PBE being called nonlocal. I think such a question as this is a good place to make other people reading my answer (who are potentially rather new to the subject) aware of that misconception so that they don't fall into that trap.