Combustion Analysis: That Voodoo That We Do

Let me start off with a confession, although it’s not an unusual one. I have only really had to worry about elemental/combustion analysis during one period in my life: while I was writing up my PhD. This is a technique that has been around for a long time in chemistry, and it sounds a bit primitive. You take a small sample of an organic compound and you burn it. Totally burn it, to where the carbon all ends up as carbon dioxide and the hydrogens all end up as water, and likewise all the nitrogen is turned into NO/NO2 and all the sulfur is turned into SO2. Then you measure the amounts of those two combustion products and determine what the ratio of carbon-to-hydrogen was in the original compound, which lets you calculate what its empirical formula must be.

I’ll illustrate that for the non-chemists in the audience. As an example, consider benzene, which is the flat aromatic ring C6H6 (molecular weight 78.1), versus pyridine, the flat aromatic ring that has a nitrogen in the ring and is thus C5H5N (molecular weight 79.1), versus cyclohexane, which is the fully saturated nonaromatic six-membered ring C6H12 (molecular weight 84.2), versus straight-chain hexane, which is C6H14 (molecular weight 86.2). If you calculate out how much of each molecular weight is carbon and how much is the hydrogens, you find that benzene is 92.3% carbon and 7.7% hydrogen by weight, pyridine is 75.9% carbon and 6.3% hydrogen, cyclohexane is 85.6 % carbon and 14.4% hydrogen, and hexane is 83.6 % carbon and 16.4% hydrogen. Note that the pyridine CH percentages don’t add to 100, because  pyridine is 17.8% nitrogen by weight as well. You can see how an accurate C/H measurement distinguishes these very clearly, and you can see the changes as you go to more saturated compounds with more hydrogens, compounds where the molecular weight is also due to added nitrogen or sulfur atoms, and compounds (none in the above examples) which might have some other element in them which will show up as “leftover” mass. 

Another key finding from elemental analysis illustrates both its usefulness and its problems. A lot of compounds form hydrates, solid (often crystalline) forms with defined numbers of water molecules associated with each molecule of the parent substance. Likewise, basic compounds can form salts (hydrochlorides, acetates, citrates, what have you) and acidic ones can be sodium salts, potassium salts, tertiary amine salts, and on and on. All of these things will of course show up in the empirical formulae and will be reflected in the molecular weight percentages. A given compound’s hydrate will show up as an unexpectedly low C and high H percentage as compared to the parent, and you can (in the ideal case) calculate to see just how many waters you have per parent molecule: one to one? One to two? Two to one? Three to two? All perfectly plausible. Maybe you made a hydrochloride salt of some amine compound and that in turn formed something like a dihydrate, which is not implausible at all either, God knows – you can calculate for that, too, of course. So while combustion analysis may sound like something out of another century, it can directly provide some information that can be rather tedious to establish by other techniques. The oxidize-and-spare-not aspect also lets you determine a sample’s purity without having to wonder about what the impurities might be: water? salt? powdered glass? pocket lint? They’ll all throw off the percentages one way or another.

But that also illustrates some problems with the technique. If you’ve made a new compound and want to establish its purity through combustion analysis, what do you do if they’re off? Maybe the numbers land perfectly if you calculate for a hydrate – peachy, but that doesn’t actually *prove* that it is one. It just suggests that as a real possibility. You might be able to dry the crap out of the compound under heat and vacuum to take the combustion numbers back to the parent values, in which case you have even stronger evidence. Or maybe the sample will immediately start shlorking water out of the air while you’re getting it into the vial to send off for combustion analysis, in which case the numbers will come back sort of in between the dry form and the hydrate, depending on how nimble the handling was along the way. There are plenty of substances that will do that. Maybe the monohydrate and dihydrate are both real substances and your sample is mostly a mix of the two? You can see where this is going: if you get a combustion analysis that’s off, you can just start plugging in possibilities and say “Ah, fine, it’s the sesquihydrate” or whatever when you find a combo that hits, and go on about your day. Did I do this when I was doing the required 

Many journals (and PhD committees!) have standards for elemental analysis of new compounds. When I was doing the required elementals for the new compounds I prepared during my PhD work, I most certainly was not above fishing for hydrates to make things come out OK. And now it’s time to talk error bars, and that’s where this new paper in ACS Central Science comes in. As it shows, the true error rates of the widely-used contract labs for this service make these cutoffs meaningless in practice. Often it’s all numbers within 0.4%, but that obviously is a wider standard for H than it is for the larger C number. Different sorts of combustion instruments have different intrinsic sensitivities and error bars for the different elements as well, but the generic cutoff doesn’t take those into account, either. No one seems to know or remember where the 0.4% threshold even came from! So here’s another confession: when I was sending my PhD compounds out for analysis and they came back wrong, I took to just taking the same damn vial and sending it right back for another analysis. If that one came back within specs, I declared victory and reported that result. Sometimes it took a couple of rounds, y’know. And from this you can see how I was not taking the elemental analyses as a scientific deal-breaker. I had mass spec data, NMR data, heck, even infrared data on all these compounds, so I was pretty flippin’ sure that they were what I said that they were. But I came to regard the elemental analysis requirement as an irritating bit of voodoo, a sort of chemistry hazing that I just had to roll my eyes about and push through somehow. I venture to guess that I was not alone in that attitude. In fact, I’m absolutely sure that I’m not. In a response to the paper cited above, Vincent Lavallo of UC-Riverside says that:

I have been practicing synthetic chemistry for 20 years and have personally experienced or witnessed, as a student/postdoc experimentalist and now as a faculty member evaluating my group and other students, many instances where elemental analysis samples have been sent and resent─sometimes up to 20 times─until the data returned matches the predicted theoretical value within ±0.4%. As a scrutinizing scientist, do you think this is a valid scientific practice? This approach is like playing the lottery.

The authors of the ACS Central Science paper also cite this recent one from a team at Vienna, examining the results from several different combustion labs. The numbers were good, but as a sequel to the above discussion of hydrates, they noted that one of their samples (cobalt acetylacetonate) came out as a 1/4 eq. hydrate when measured on the in-house instrument, and had moved on to a 1/2 eq. hydrate in transit to the mail-in labs. But even so, this group noted that the literature elemental analysis values are (as a whole) still too precise to be real, given the actual error bars and variations between different instruments. 

The results impressively show that typical deviations from theory are in the range of 0.05–0.20% even when high purity commercial compounds are analysed. Just in few cases all values of a compound are ≤0.05%. Notably, in ∼40% of the analysed elements already the standard deviation from triplicate measurements is ≥0.05 (Table 1 and Tables S1–S5). This means when the same sample is measured several times at the same instrument, a difference of ≥0.05% between the individual measurements is not unlikely. Therefore, it is hard to believe that a whole set of compounds in a publication shows perfectly fitting values with ≤0.05% for all elements.

So there seems to be a general whistle-blowing on the way that we’ve been doing elemental analysis all these years, and a realization that perhaps we would be better off if we all stopped kidding ourselves. What should we be doing instead? Well, if we’re going to still require this purity check, the Vienna group recommends that far more data be provided: the instrument used, the raw data from the analysis, the error bars for each element, etc. The ACS Central Science authors recommend that we at the very least establish different error tolerances for the different elements (nitrogen in particular seems to vary more in their data). For my own part, I think of myself back in grad school in early 1988, muttering to myself as I send off another batch of compounds for combustion, and I wonder what I would have thought about everyone finally throwing up their hands in 2022!