Let me add my 2 *pence* worth, as the Manchester team was mentioned too. I agree with Calvin that a true 'ab initio' EXAFS black-box for a sample about which nothing else is known is probably a non-sensical proposition. As with so many other (any other?) technique that probes molecular level properties one needs additional information about the sample, or more gerenally, 'prior knowledge'. Now, with XAFS we have a technique that is really only let loose on samples (if not the experimentalist then the beamtime committees will usually see to that...) that have already been characterised to quite some level of detail. Even during in situ measurements on dynamic systems (reactions, phase transitions, in vivo, etc), where the nature of the sample is more open-ended, the possibilities are quite restricted. But the range of possible elements in the sample is practically always limited (at least within the concentration ranges that can be probed by XAFS), there are also rules about reasonable ranges for bond lengths out there, phase diagrams, etc. All of these constrain the possible parameter space quite significantly - and that is what is currently done through the human operator. The conceptual challenge for the design of any 'XAFS black box' is to find a way to integrate this prior knowledge into the analysis process. I, as a physical chemist, sometimes call the challenge 'How to teach a computer chemical intuition'. Bond length rules, or more generally, any knowledge from inorganic chemistry textbooks - there is no principal reason why all this could not be coded into a database for a decision-making system, which would then have to be coupled to a generic EXAFS analysis interface that allows people to constrain a search according to what they already know about their samples. A Herculean job before it becomes elegant... Also, let's not forget the XANES. In principle that is information that you can almost always get 'for free' with any EXAFS measurement. There is really a lot of information to be mined through proper XANES analysis - but we still have a lot of work to do to get there ... Good news I guess. Keeps us busy for some years to come! So that makes roughly 8 cents now (1p = 2c, approximately, at the moment at least). -- Sven L.M. Schroeder (mailto:s.schroeder@manchester.ac.uk) School of Chemical Engineering and Analytical Science (CEAS) & School of Chemistry The University of Manchester PO Box 88 Sackville Street Manchester M60 1QD United Kingdom http://www.slmslab.info Tel +44 (161) 306 4502 Lab +44 (161) 306 4486 Fax +44 (161) 306 4399 Offices: School of CEAS: Room C17 (Jackson's Mill) School of Chemistry: Room E3 (Faraday Undergraduate Block) DISCLAIMER The views expressed within this message are those of the sender, not those of The University of Manchester or one of its Departments. While all emails and attachments are scanned for viruses before sending, we cannot accept any responsibility for viruses, so please scan all emails and attachments. This email is intended for the addressee only. If you are not the intended recipient, please notify the sender and delete this email immediately. END OF MESSAGE
-----Original Message----- From: ifeffit-bounces@millenia.cars.aps.anl.gov [mailto:ifeffit-bounces@millenia.cars.aps.anl.gov] On Behalf Of Scott Calvin Sent: 30 July 2006 07:21 To: XAFS Analysis using Ifeffit Subject: Re: [Ifeffit] Other programs than FEFF - GNXAS and MXAN
Well, I have some time to waste, and since my name was just mentioned...
I think a black box is a wonderful idea for cases where the space of possible solutions is very limited. At the same conference that the UK group presented, Wolfram Meyer-Klaucke gave a very nice talk on a black-box system for determining protein structure around an active site. The talk was opposite the UK group's, so unfortunately no one saw both.
In any case, that's a perfect situation for automation: there are a very limited number of ligands that could be present, and their structures are very well understood. It's much more than fingerprinting, since the combination of ligands might never have seen before. Their system even allows for the ligands to be at slightly unusual distances. But it only works because the biochemistry is already pretty well understood and quite limited. Although a very different system, I suspect the UK automation of supported metal catalysts has similarly limited scope; neither system would probably work very well when fed data meant for the other!
I think there's sometimes a wish for a system that acts like a Star Trek tricorder: stick any spectrum in, and the computer can say "it appears to be an oxide with coordination number 6." In my opinion, that kind of system will never be developed, because there just isn't that much information sitting in the EXAFS. (Some of you new to EXAFS may be puzzled by that--it doesn't sound like very much information at all. But if you're going to allow me the space of all possible structures along with less-than-perfect data, it's hard to distinguish disorder from coordination number changes, for example, and it's in turn hard to distinguish true disorder from splitting below the resolution of the data.)
In other words, current experts in EXAFS analysis don't act as black boxes to the outside world. If <pick-your-favorite-expert> were brought a spectrum and asked "OK, tell me the structure," the expert would immediately start asking questions to gain additional information (or would say "no," and walk off in a huff). If experts don't act as black boxes, then neither can a computer.
OK, with my two cents added to Bruce, that now makes four...
--Scott Calvin Sarah Lawrence College
Ah! The "black box" discussion. That's a fun way to waste lots of time. ;-)
The physics has its complicated parts, but I suspect that we have a sufficient understanding of the problems. There are many applications where a black box is reasonable to consider and probably would even work pretty well. At the recent XAFS conference, the group from Manchester, UK presented a high-throughput scheme that involves automate processing of larfge quantities of data. They seem to get good results with minimal human intervention. And Harald Funke gave a really neat talk about Feff-based wavelets that could be a very useful approach to a first-shell black-box.
There will always be a large part of exafs analysis that falls well outside the scope the black box. The sorts of crazy fits published by some of the frequent contributors to this list (I am thinking specifically of Scott Calvin and Shelly Kelly) will always defy automation.
Well, that was my US$0.02 worth... B
_______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit