>There seems to be a general agreement that you get best results with >calculating both the geometry optimization as well as the GIAO shifts on >B3LYP/6-31G(d) level, but I trust whatever Henry suggests :-) >With my own Gaussian calculations, I got excellent results for rigid and >unpolar compounds, just as Barone suggested in his paper. >
Dear all; being off-line yesterday, I am picking up the thread here (which may be superceded by later posts). The historical line might be of interest. At the San Francisco ASC, I along with about 500 people piled into the lecture given by James La Clair; subsequently about 3 weeks ago, I got to read Scott Rychnovsky's refutation of the Hexacyclinol original structure, using geometries optimised at HF/3-21G and GIAO done at mpw1pw91/6-31G(d,p) (the Bifulco protocol!). He obtained a mean absolute deviation of 1.8 ppm for the 23 13C shifts, and a maxi absolute deviation of 5.4. This was generally accepted by the community as a convincing proof of the new structure, aided by biosynthetic evidence. It was subsequently confirmed by X-Ray. Scott did three things, two brave, and one unnecessary 1. He picked a "high energy conformation" for a side chain as giving the better fit. The X-ray vindicated this 2. He transposed two assignments 3. He "corrected" the predicted shifts by LS calibration against the obs shifts, and then obtained the 1.8 ppm MAD from the "corrected" predictions. The article makes for an interesting read in this regard; we can perhaps forgive the liberties he took with statistics because he is a "synthetic organic chemist". A quick read of this paper immediately suggested improvements, which I tried out 1. To use 6-31G(d,p) geometries. 2. To apply a solvation correction for chloroform. This has the greatest impact upon carbonyls, often improving them by 3-5 ppm. 3. To dispense with all "correction factors". This yield a MAD for hexacyclinol of 0.9 ppm, with a max deviation of 2.8 ppm. I wrote to Scott with these results, and he agreed to all the points. We are planning a number of projects from this. It has transpired that hexacyclinol was a "lucky" choice. Systematic errors started emerging from the above protocol 1. The method is sensitive to conformation. Positively, it can be used as a real conformational analysis tool, but negatively, it can mean a LOT of work. This is of course generally recognised. 2. The 6-31G(d,p) basis is also a lucky one; a great deal of cancelation of errors occurs. Thus ramping up to aug-cc-pVTZ is a disaster (because some errors are removed, leaving no cancellation!) 3. We find the aug-cc-pVDZ is better, especially for esters, amides etc. But it can be a pathologically bad SCF converger! I could go on and on, but the above should suffice, if only to set out how "science really works". The above was achieved by active collaboration between only two people, and it has to be said an imperfect searching of the literature by one of them! The result is a practical (rather than theoretically sound) protocol. I agree totally that if we open this up, then the improvements could both come more quickly, and hopefully also be more soundly based theoretically. -- Henry Rzepa. +44 (020) 7594 5774 (Voice); +44 (0870) 132 3747 (eFax); [EMAIL PROTECTED] (iChat) http://www.ch.ic.ac.uk/rzepa/ Dept. Chemistry, Imperial College London, SW7 2AZ, UK. (Voracious anti-spam filter in operation for received email. If expected reply not received, please phone/fax). ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2005. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ Blueobelisk-discuss mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/blueobelisk-discuss
