Today I have my brain split.
P., with whom I am to work on the semiconductors project, starts work today. This week, we are supposed to learn how to use the HPLC, whatever gas chromatography equipment is reserved for our use, get our lab area set up in general (we have to keep completely separate glassware, for instance), and set up our lamps for running experiments with the photocatalyst.
But I’m not done with the Antartica project yet. Jenn, and I may very well be running a decidedly large experiment this week, doing a photobleaching experiment in triplicate, then basically running three sets of analysis on the samples we draw out at various timepoints. The photobleaching takes 48 hours solid, and running the 90 (for some reason Dr. M said 180, but I can’t determine a reason to double my number, unless we do “darks”, or un-photobleached samples, too) samples on the UV/Vis, which is another week of work, because the machine is slow. And the lamp setup to be able to run the photobleaching experiment will take a day or so, as there seem to have been some problems with steady output the last time an experiment was run. Meaning I get to learn some actinometry.
The results from running the samples through the UV/Vis will have to be run through a sexy stats package (I haven’t gotten to play with R yet, but that will most likely be what I use) and fit to the empirical model that we’re having trouble with. Model fit and residuals will be analyzed. Insert the learning curve for me to learn the language of R.
The whole point is to determine if we can shorten the entire process of gathering samples and analyzing them. It looks like the best way to get a reliable estimation of error on one of the values we calculate is to take samples and run the photolysis experiment in triplicate. See above about the week it will take to analyze 90 samples; this isn’t practical in situ. So the idea is, what if we just take eight wavelengths in the spectrum we analyze instead of the entire range from 500 nm to 200 nm, since our model bias and machine drift are damned constant and we’re getting correlation coefficients of 0.998 when curve-fitting? If we can get comparable results, then running the experiment in triplicate becomes much more manageable.
Of course, I think this still leaves us with the dilemma of estimating the model error and machine drift. The machine seems to be “drifting” in an oscillating function. Now, on top of that, and in comparable magnitudes, we have an oscillating function that is our model bias. (I say “oscillating fuction”, but I mean that our residual plots are one big sine curve [the model bias] with jumps and incontinuities from the machine drift.) Now, the machine drift makes me think of jumping double-dutch: since the lamp and detector of the UV/Vis are always on, the machine is always drifting in this oscillating manner. When we put in a sample and run it, we’re essentially “jumping in” at some unknown phase shift of the drift function, just like jumping double-dutch. We had some ideas for experiments to run to erradicate this (so we’d be left with only the model bias, which I think may be able to be stamped out with methods similar to those involved in time series models, leaving us with measurable random error) that we tossed at the awesomely cool Regressions Guy in an e-mail Friday afternoon. We could very easily be completely wrong, since I have no experience with regressions and the time series chapter in the textbook certainly didn’t cover model bias evaluation as an appropriate application.
Once we determine our method, those will need to be implemented in the program I’m sitting on, which could mean anything from me needing to change a good deal of the input algorithms (taking in the triplicate values and processing them accordingly) to needing to change the model to one with a second variable (two-variable optimization algorithms, anyone?). Insert another week of me writing algorithms and implementing options and testing consistency with the stats program we use. And I mean, most likely, a 40-hour week.
But the semiconductors project isn’t just us–there’s another school (hell if I remember what it is) involved in this initial setup to perform the photocatalyst experiments, meaning we need to get on the ball. Now, given how we (Dr. M, P., and I) will have to make sure some of these machines can be used at all, and for our purposes in particular, that we may be bumping elbows with at least one other research team, that I, at least, have no practical experience on these machines (I think P. has taken the pre-requisite courses), and that Dr. M has been known to drive me absolutely batty with her very loose style of organization and tendency to just disappear for hours, this week of setup for the photocatalyst side of things could be harrowing.
Then there’s the fact that, as of last Wednesday, I think, my grant money hadn’t come in yet, and the next pay period should start drawing from those resources.
I feel sorry for Dr. M, who’s managing two teams on completely different projects; the workload on Jenn and my project should have dropped to the point where she’d need only weekly updates on our progress, but it seems to have grown exponentially, with Jenn needing to re-run a solubility enhancement experiment, then the above nastiness with the photolysis experiments. I would feel sorry for myself, who’s struggling just to learn enough to be of assistance to anyone and who just had the knowledge requirements triple, but I knew what I was signing up for.
I suspect something will give, or else the teams will have to start working long hours. This wouldn’t be so bad for me, since I don’t mind being at school (away from the “comfort” of home), I’d have Internet, and I could get more work done (it’s hell to be held up by a simple need to look up something, but not have the ability to do so), but there could be increased grumpiness and tension in the workplace. But given that another team under another prof works from 05:30 to 17:00 or 18:00 six or seven days a week, and that we all signed on for salaried positions knowing we’d need to work “until it got done”, I couldn’t complain.
But I think something will give first, because I don’t think Dr. M wants to work hours like that, and she would have to put in some extra time just to keep up in managing us. Of course, things might turn out not to be so bad, either (Dr. M isn’t in yet for the day): Jenn and I may not need to run the long experiment, or the Regressions Guy could have had a big “ah-HA” over the weekend, or the machinery P. and I will use could be in a wondrous state and ready to go once we knock the cobwebs off.