search

nwchemgit / nwchem

NWChem: Open Source High-Performance Computational Chemistry
http://nwchemgit.github.io
Other
493 stars 160 forks source link

IPCCSD Gave Different Excitation Orbitals #42

Closed xiongyan21 closed 5 years ago

xiongyan21 commented 6 years ago

Dr.Faklerus raised a question on the NWCHEM forum, saying failing to run the QA for f2 IP-EOMCCSD calculation. I find there is no prolem to finish the calculation using the original input to get the results identical with those in QA test. I have just installed the updates of the operating system, and have not recompile NWCHEM6.8, and now the calculation gave different excitation orbitals. ... Excited state root 1

Excitation energy / hartree = 0.554880518711326

              / eV      =       15.099073582827362

largest EOMCCSD amplitudes: R1 and R2

Singles contributions

8a   (alpha)       -0.9552531110

Doubles contributions

10a (beta ) --- 5a (alpha) 7a (beta ) -0.1018834489

10a (beta ) --- 6a (alpha) 7a (beta ) 0.1101299892

Parallel integral file used 4 records with 0 large values

Task times cpu: 24.8s wall: 42.4s

... Another run gets ... Singles contributions

8a   (alpha)        0.2922556835

9a   (alpha)        0.9094477014

Doubles contributions

10a (alpha) --- 6a (alpha) 7a (alpha) -0.1058335725

10a (alpha) --- 7a (alpha) 6a (alpha) 0.1058335725

10a (beta ) --- 6a (alpha) 7a (beta ) -0.1444066816

I will recompile it and see. ip.log ip1.log

ip2.log ip3.log ip4.log ip5.log

jeffhammond commented 6 years ago

I have upgraded the operating system to Ubuntu18, and have not recompile NWCHEM6.8, and now the calculation is not stable

Don’t do this. Recompile binaries when you upgrade the OS.

xiongyan21 commented 6 years ago

On macOS High Sierra, they are also different. ip2.log ip3.log ip1.log ip.log

xiongyan21 commented 6 years ago

Recompilation of NWCHEM6.8 on Ubuntu makes the double contribution different. ip.log ip1.log

xiongyan21 commented 6 years ago

Recompilation of NWCHEM6.8 on Ubuntu makes the double contribution different. ip.log ip1.log

I have asked precompiled GAMESS for MAC to do the same calculation here with IP-EOM3A, i.e., up to 3h2p excitations. I used mp2 to optimize the geometry and changed the basis set into aug-cc-pVTZ to obtain IE, and it could be 15.749 eV. It is wonderful for GAMESS to begin with an RHF calculation and then automatically enter an ROHF one. GAMESS IP-EOM2 including up to 2h1p excitations could give the IE of 15.619 eV, and NIST Chemistry Webbook gives vertical IE of 15.70** eV.

With the original input, the first excitation energy from NWCHEM is around 0.6 eV deviating from experimental data. Perhaps this is not IE, but IE would be obtained based on it.

GAMESS ROHF IP-EOM2 could give IE of F atom 17.35944 eV using aug-cc-pV5Z, compared to that of 17.42282 eV on NIST Chemistry Webbook.

ROHF treatment of radicals and other open-shell systems is one of the great challenges in computational quantum chemistry, and it is very likely that the IP calculated employing IP-eom2 has a large discrepancy, or is completely wrong when compared with experimental data, especially for a system with significant multireference features.

The successful calculation of such a situation in GAMESS lies in its employment of the artful and ingenious IP-EOM3A and EA-EOM3A to provide results. I did this way and obtained IE of CH3 whose geometry mp2 optimized of 9.67 eV，when the number of active orbital was one and the basis set was aug-cc-pVQZ, definitely acceptable when compared with experimental 9.84+-0.01 eV on NIST Chemistry Webbook.

I think we should make things physical and rational although the deviance might be larger when compared experimental data.

The open-shell EA and IP coupled-cluster programs in GAMESS were pogrammed by Dr. Gour, Prof. Piecuch, and Prof. Wloch, evaluated and assembled by Prof. Gordon, Dr. Schmidt and other doctors in Prof. Gordon's Group or Ames Laboratory.

xiongyan21 commented 5 years ago

Can the motivation of the different excitation orbitals be enlightened by the article of Prof. Gordon, et al.: Can Orbitals Really Be Observed in Scanning Tunneling Microscopy Experiments?

xiongyan21 commented 5 years ago

GAMESS can directly give IP. Is the first excited root obtained by NWCHEM here the first IP？ Very Best Regards!

xiongyan21 commented 5 years ago

GAMESS can directly give IP. Is the first excited state root obtained by NWCHEM here the first IP？ Very Best Regards!

xiongyan21 commented 5 years ago

I can verify the first excitation energy obtained is the first IP using NWCHEM because when I changed the keyword IPCCSD into CCSD in TCE, the obtained first excitation energy is much lower, i.e., 4.691836421578004eV, which is also significant larger than that in an article, thus the IP and the first excitation energies obtained by NWCHEM are unsuccessful for F2 when compared with experimental data and the multireference CI method, respectively.