Column: Bench monkey

Dylan Stiles is bedazzled by computational chemistry

As I’m tapping out this column on a large, shiny Macintosh, an amazing program is running in the background, churning out calculations in an attempt to justify some of my bench results. I’m running the deluxe edition of Spartan ’06- an application with all the bells and whistles to let you analyse just about anything to do with the periodic table. 

I’m an experimental organic chemist, with flasks and beakers being my tools of the trade, so this is all pretty new to me. But I must admit that I’m pretty impressed by this program. 

The last time I played around with serious chemistry software was a good five years ago. Since then, Moore’s law has held up and computers have come a long way. In fact, this particular version of Spartan  won’t even run on a Mac that’s more than a couple of years old - we needed to buy a new ?1000 computer, plus the ?600 software package, to run the beast. 

And my calculation has been occupying its full resources for over a day, with the processor running in the red zone and the computer hot to the touch. Clearly it’s thinking very hard. 

Blind faith 

The fact that this extremely fast computer is making so much effort to produce its data gives me blind faith in its reliability, and I’m enthusiastic about getting results that are in line with my predictions. But I can’t shake the nagging sensation that I don’t actually know what I’m doing. 

There’s a gaping hole in my scientific knowledge about computational chemistry. Obviously a huge field unto itself, I’ve never taken a single class on the topic. My understanding of the mathematical models that describe molecules pretty much stops at the Schr?dinger equation. 

Exhibit A: Spartan  tells me that it’s computing a ’Hessian’. This means nothing to me, and Google’s information¹ that a Hessian matrix contains ’the matrix of second derivatives of energy’ is no help. Is it better if my Hessian is big or small? I have no idea. 

The responsible thing to do would be to take a class on computational chemistry before any bad habits get engrained. Or at least seek the guidance of an expert. Right now, I’m basically trying to teach myself to play golf from first principles, with mixed results. 

While the documentation provided with Spartan  is incredibly helpful, it would take a while to make it through both the 400 page-long tutorial and the 800 page-long guide to choosing the right computational method.² And although Spartan  is as intuitive as it could be, given its capabilities, I feel like I’m sitting behind the steering wheel of a space shuttle with only The User’s Guide to Flying to the Moon  at my side. 

As a nervous driver, that means I feel compelled to check my technique by practising on known entities and comparing my results with the facts. This has been done before, quite extensively in the behemoth guide provided by Spartan  and also by third parties.³ The reassuring conclusion is that the answers from these calculations are often quite reliable indeed.  

You can’t argue with results like that - so while I may not understand exactly how these programs work, I trust them the same way I trust an airline pilot to get me where I need to go. I could just use some flying lessons. 

Dylan Stiles is a PhD student based in California, US