|
|
★ 小木虫(金币+0.5):给个红包,谢谢回帖
看了一下GMTKN30原文,貌似作者对渐进行为的指责主要有三处
1."It has been noted that these two double hybrids still lack an asymptotically
correct description of long-range London-dispersion effects (aC < 1), although the inclusion of the nonlocal PT2 part already leads to a qualitatively better description compared to common DFs. Therefore, it was suggested to combine the functionals with an empirical London-dispersion correction (DFT-D).24”
2. which is worse with B3LYP due to the wrong asymptotic behavior of the larger
(compared to B2LYP) GGA part
3. Note, however, that this only holds for small- and medium-sized complexes as XYG3 misses asymptotically about 57% of the dispersion energy due to an effective aC of 0.43.
第二处实际是GGA部分的。就我个人的理解,作者对渐进行为的批评针对的是色散能大小继而对应C6系数。似乎这里的任何一种double-hybrid泛函的计算框架都不是完善的,精度更多的是一些参数拟合的结果。
关于"专门的色散校正项" ,MP2两个闭壳层分子的解离极限是uncoupled hartree-fock的C6
SZABO, A., and OSTLUND, N. S., 1977, J. chem. Phys., 67, 4351.
G. CHALASINSKI M. M. SZCZESNIAK, MOLECULAR PHYSICS, 1988, VOL. 63, NO. 2, 205
按Szabo等人的推导, 只要有MP2的关联能形式,用各种DFT轨道都可以得到对应uncoupled-KS的C6. 无论是uncoupled hartree fock还是各种近似的交换关联泛函,没有一种是完美的。在double-hybrid泛函形式下重新参数化或者增加一个色散修正项都可以改进C6系数。作为一种拟合方法,很难保证在对于拟合库外的分子的计算一直是准确的。而且这些都是基态,如果计算激发态,是不是要一个激发态一种C6?
关于XYG3使用B3LYP轨道计算PT2
"In their original XYG3 paper, the authors argue that the hybrid-GGA part of B2PLYP (i.e., B2LYP) does not employ 100% of DFT correlation" VS B2PLYP etc。这里的double-hybrid DFT没有一种是严格在KS-DFT下的Levy的微扰展开相关泛函的框架下进行的
A. Gorling and M. Levy, Phys. Rev. B, 1993, 47, 13105
A. Gorling and M. Levy, Phys. Rev. A, 1994, 50, 196
因为他们都使用了non-local Fock exchange potential,而KS-DFT的potential是严格局域(local)的。
这里的potential指v_xc:=δE_xc[ρ]/δρ, local指在坐标表象可以写成Dirac delta函数的形式=f(r)δ(r-r')。
虽然KS-DFT不是所有的DFT,但B2PLYP/XYG3....用Fock exchange,SCF收敛后用PT2的计算更多是一种技术上的妥协,看上去没发现完整的理论基础。(不过如果有兴趣可以尝试一下把这种'理论'给justify出来^_^) 反正最后拿不同的泛函对着实验数据拟合参数(....'XYG3 misses asymptotically about 57% of the dispersion energy due to an effective aC of 0.43.'),对于某些分子,某种泛函的精度可能好一点
总之,个人不认同作者的推理
或者“MP2在长程只占有部分成分”另有所指...
[ Last edited by fichte on 2011-5-6 at 20:43 ] |
|