Embrace failure?

Seth Olsen alerted me to a thought-provoking article in Wired about the importance of failure in science.

Think before you calculate, measure, fabricate,....

What will be your New Year's resolution for your research in 2010?

Write more papers? Get more students? Apply for more grants? Get more lab space? Get a paper published in Nature or Science? Learn a new technique?

I think mine is going to be:

Spend the first half hour of each day thinking and writing in a notebook about the important science questions I am interested in and want to try and answer. And, specifically coming up with multiple alternative hypotheses and devising ways to distinguish them.

Where did this come from?

Previously I wrote a post about a beautiful Science paper about Strong Inference by John R. Platt. He references a book he published in 1962, The Excitement of Science. Unfortunately, it is out of print and only two universities in Australia have a copy in their libraries. I got a copy on interlibrary loan and just finished reading it. I have scanned a copy of chapter 7 and chapter 8, which I found the most helpful and challenging.

Electronically driven hula dancers

Over the 25 years Robert Liu [from the University of Hawaii!] has emphasized that the

"hula twist" is ubiquitious in photoisomerisation reactions [where after a molecule absorbs a photon it undergoes a structural change] and is beautifully summarised in the figure below.

This short review by Liu and Hammond [whose address is Aloha, Oregon!] documents this and also argues that the hula twist is driven by steric considerations because it is "volume conserving".

However, it turns out that this geometrical change can be preferred purely by an electronic mechanism and does not require steric hindrance. Seth Olsen and I show this [amongst many other results, some of which I have discussed in a previous post] in a paper just published in Journal of Chemical Physics.

We refer to the "hula twist" as disrotatory motion and discuss it briefly on page 12 of our paper.

Trust, but verify

Earlier this year a Nature paper reported the data below [black squares with error bars] for the spectrum of high-energy cosmic-ray electrons. The peak was interpreted as evidence for 500 GeV particles (dark matter) predicted by generalisations of the standard model that include extra dimensions.

However, more recent data [shown as red points] from NASA's Fermi Gamma-ray space telescope was recently published in PRL. The data has much better statistics and shows no peak. More details can be found here.

This is a good cautionary tale. There is no substitute for two or more independent experiments to test a hypothesis.

I don't think Ronald Reagan was a good U.S. president, but his signature phrase "Trust, but verify" has merits.

Talks that go pear shaped....

Every now and then you go to a seminar which goes horribly wrong for the speaker. Someone asks a question that the speaker answers poorly or cannot answer. Then other people start asking questions or offering critical comments and it gets worse.....

Why does this happen? How can the speaker prevent it?

I think it may be because the speaker violates the important principle:

Never offer undefendable ground.

i.e. do not make claims that you cannot back up

Speakers will sometimes make claims that are not necessary for the actual talk but will irritate members of the audience, particulary senior people. I think students who "parrot" lines they have learnt from their advisor about justification for their work.

A random set of sample claims which you may hear variants of include:

-our results will allow the design of new materials

-silicon based electronics has no future

-density functional theory cannot described electronic correlation effects

-molecular dynamics simulations of biomolecules tell us nothing

-the Hubbard model oversimplifies the true Hamiltonian

-BEC's allow us to tune parameters in a manner that is not possible in traditional condensed matter systems

-quantum information processing is going to revolutionise computing

-our theory agrees with all the experimental results

-everyone elses theory is wrong

So if you don't want your talk to go pear shaped, don't claim anything you wont be able to defend.

How big a Hilbert space do you need?

How big a Hilbert space do I need to describe the electronic properties of a molecule?

Specifically, suppose in the molecule there are N valence electrons.

One must decide then how many spatial orbitals are required and how many Slater determinants? This issue is brought out in this review paper. Benzene is an illuminating case. McWeeny shows that a "brute force" approach based on molecular orbital theory requires hundreds and sometimes thousands of Slater determinants to obtain results of comparable accuracy to that obtained with a valence bond description.The latter uses just six localised orbitals and two Slater determinants (corresponding to the two Kekule structures).

Why does this matter? First, the priority of chemical insight favours the valence bond description over the "black box" approach embodied in the molecular orbital theory approach. The issues are described nicely in this Nature paper, which emphasises that the delocalised molecular orbitals are physically misleading. Second, computational efficiency may be more likely to be obtained with the simpler description.

A key question is whether one can codify these issues in a systematic way (perhaps with ideas from quantum information theory) to develop quantitative criteria to decide what is the minimal number of orbitals and Slater determinants.

Thanks to Anthony Jacko for providing the cartoon.

When business people think they can run a scientific organisation...

Articles from Nature, ScienceInsider, and The Age newspaper summarise recent problems concerning the strained relationship between the board (chaired by a corporate lawyer) and scientists at the Australian Synchrotron. The International scientific advisory committee has resigned in protest.