Green Solvents for Analytical Chemistry: The Ranking that Doesn't Exist

The previous post in this series was about removing the solvent entirely. Solid-phase microextraction needs no extraction solvent at all, and that absence is the appeal. It seems to end an argument before it starts: if the solvent is the problem, the greenest choice is none.

But most methods cannot work that way. Liquid-liquid extraction is still the backbone of regulated environmental analysis, and it runs on solvents most guides would rather you did not use. So the question comes back, and with it the tool everyone reaches for first: the solvent selection guide[1]. Open one and you get a tidy table: Recommended, Problematic, Hazardous. It looks like a ranking.

It is not a ranking. It is a ranking under a set of assumptions nobody prints on the card, and when you line up the major guides against each other, they do not agree. Nobody defends benzene. Everything interesting happens in the middle of the table, where the guides part ways - and where the solvents we actually use everyday live.

Why the guides disagree

The disagreement starts with scope. A solvent selection guide scores the solvent in the bottle: GHS hazard classifications, boiling point, aquatic toxicity, regulatory status, something about how it is manufactured. The CHEM21 guide (Prat et al., 2016)[1], the GSK guide, the ACS GCIPR Pharmaceutical roundtable guide[2] all work this way. They answer question about a chemical. Method-greeness metrics answer a question about a procedure: volumes, energy, waste generated, sample throughput. These are different questions wearing the same word, and a lot of confusion in the green-chemistry literature traces back to that collision.

Even between guides scoring the same thing, the answers differ, because the weights differ. CHEM21 scores safety, health, and environment on separate 1-10 scales and then lets a panel of experts argue[1]. That argument is not a footnote, it decides the outcome. Methanol sits in different categories on different drafts of the guide depending on how the panel weighed flammability against everything else. The AGREE family of matrices takes a different route: it hands the weighting to you[3]. You assign the criteria weights yourself, which makes the tool flexible and makes the answer yours. A score that encodes the assessor's values is not a measurement.

Then there is the boundary question: Does "environment" cover solvent manufacture and disposal, or just bench exposure? Does a bio-derived solvent count as green if it competes with food crops for land? The guides answer these silently, in their assumptions rather than their headings, and two guides with different answer will rank the same solvent differently with no visible reason why.

And no guide reproduces another's number. There is no ground-truth ranking to converge on, so there is no reason to expect convergence. That is the sense in which the ranking does not exist.

The metric zoo

The methods side of the field grew its own tools, one per complaint about the last one.

NEMI came first: a set of pictograms, green-yellow-blue per criterion, no aggregation at all[4]. The Analytical Eco-Scale (Gałuszka et al., 2012, [5]) went the other way — penalty points subtracted to a single 0–100 number. GAPI (Płotka-Wasylka, 2018, [6]) drew the whole analytical procedure, collection through determination, as a five-pentagram pictogram; ComplexGAPI (Płotka-Wasylka & Wojnowski, 2021, [7]) extended it backwards to include the pre-laboratory stages — reagent production, transport, storage. AGREE (Pena-Pereira, Wojnowski & Tobiszewski, 2020, [3]) mapped the twelve principles of green analytical chemistry onto a weighted clock face, and AGREEprep (Wojnowski, Tobiszewski, Pena-Pereira & Psillakis, 2022, [8]) rebuilt the idea specifically for sample preparation, with ten criteria and user-set weights. The AMGS calculator, pharma-driven and separation-focused, rounds out the set[9]. Sajid & Płotka-Wasylka's 2021 review is a good single map of the whole zoo[10].

Each tool exists because the previous ones missed something, which is an honest confession that no single score captures "green." That is fine. The zoo is a feature, as long as you know which question each animal answers - and the failure mode is using a guide's answer to a method question, or vice versa.

What a guide score cannot know

Here is the part that matters for anyone who actually runs extraction.

Dichloromethane is hazardous on every card, ind it is still the workhorse extractant in EPA-syle liquid-liquid extraction. Method 3510C species esrial extraction with methylene chlorine, and it does so for reasons a guide cannot see: high density for clean phase separation, broad analyte coverage, decades of validation behind every recovery. The guide ranks the solvent. Only the method developer can rank the trade.

Two facts do most of the work in that trade. First, volume beats identity. Ten milliliters of a "problematic" solvent usually has a smaller footprint than a hundred milliliters of "recommended" one, which is why miniatuarization moves the greeness needle more than swapping bottle A for bottle B - and why AGREEprep's criteria weight volumes so heavily[8]. A greener extractant with poorer recovery pushes detection limits up, forces larger injection volumes or more replicates, and consumes more of everything else to compensate. A method that misses the analyte is not green; a failed analysis wastes the sample, the solvents, the instrument time, and the operator's day. The eco-point of a method has two terms, and the guides only see one.

Score one extraction two ways

Take a concrete case: extracting trace organic contaminants from water. Three plausible routes. Classic DCM liquid–liquid extraction, 60-odd mL per sample. Solid-phase extraction with a small-volume methanol elution, a few mL. SPME, which we already covered, with none.

Score the solvents on a CHEM21 card and DCM lands in the problem column, methanol in the recommended one. SPE looks better, SPME perfect. Score the whole procedures with AGREEprep instead, and the ordering stays roughly the same - but for different reasons. The DCM method loses points for volume and waste, not because of the solvent's identity; the SPE method gains on volume but pays on the sorbent and the multi-step workflow; SPME wins mostly by making most of the criteria moot. No comparison here should be treated as precise: no head-to-head study scoring the same extraction family under two different metrics turned up in the literature search for this post.

The qualitative point survives the hand-waving. Ask "which solvent is greener?" and you get one answer. Ask "which extraction is greener?" and you get another, and the difference between them is the whole argument. The honest output is the score and its assumptions, reported next to the validation data - the same discipline this series argued for in ML prediction accuracy. Scores are models of reality, not reality.

What to do with this

Use the guides for what they are good at: shortlisting. When you are choosing a new solvent, a selection guide is a fast way to exclude the worst options before you think harder. Use the method metrics for comparing procedures, where volume, waste, and energy actually differentiate. Then let the validation data decide, because a method that fails is green in no useful sense.

Report the assumptions. Which metric, which weights, which system boundaries. A greenness score printed without its assumptions is a marketing number, and journals are starting to expect the details.

The real frontier is shrinking the problem rather than re-ranking solvents: miniaturization, solvent-free sampling, in-situ extraction. And past that sits an open question that nobody has solved: a genuinely analytical greenness ranking would be performance-weighted, combining greenness with extraction efficiency and detection limits in one defensible framework. Nobody has built it, and the guides' disagreements over something as simple as where to draw the boundary suggest why. Until then, the ranking that does not exist stays nonexistent - and the methods that do exist have to be judged case by case.

References

[1] Prat, D.; Wells, A.; Hayler, J.; Sneddon, H.; McElroy, C. R.; Abou-Shehada, S.; Dunn, P. J. CHEM21 Selection Guide of Classical- and Less Classical-Solvents. Green Chem. 2016, 18 (1), 288–296. https://doi.org/10.1039/C5GC01008J

[2] Diorazio, L. J.; Hose, D. R. J.; Adlington, N. K. Toward a More Holistic Framework for Solvent Selection. Organic Process Research & Development 2016, 20 (4), 760–773. https://doi.org/10.1021/acs.oprd.6b00015

[3] Pena-Pereira, F.; Wojnowski, W.; Tobiszewski, M. AGREE─Analytical GREEnness Metric Approachand Software. Analytical Chemistry 2020, 92 (14), 10076–10082. https://doi.org/10.1021/acs.analchem.0c01887

[4] https://www.nemi.gov/home/

[5] Gałuszka, A.; Migaszewski, Z. M.; Konieczka, P.; Namieśnik, J. Analytical Eco-Scale for Assessing the Greenness of Analytical Procedures. TrAC Trends in Analytical Chemistry 2012, 37, 61–72. https://doi.org/10.1016/j.trac.2012.03.013

[6] Płotka-Wasylka, J. A New Tool for the Evaluation of the Analytical Procedure: Green Analytical Procedure Index. Talanta 2018, 181, 204–209. https://doi.org/10.1016/j.talanta.2018.01.013

[7] Płotka-Wasylka, J.; Wojnowski, W. Complementary Green Analytical Procedure Index (ComplexGAPI) and Software. Green Chem. 2021, 23 (21), 8657–8665. https://doi.org/10.1039/D1GC02318G

[8] Wojnowski, W.; Tobiszewski, M.; Pena-Pereira, F.; Psillakis, E. AGREEprep – Analytical Greenness Metric for Sample Preparation. TrAC Trends in Analytical Chemistry 2022, 149, 116553. https://doi.org/10.1016/j.trac.2022.116553

[9] https://acsgcipr.org/amgs/

[10] Sajid, M.; Płotka-Wasylka, J. Green Analytical Chemistry Metrics: A Review. Talanta 2022, 238, 123046. https://doi.org/10.1016/j.talanta.2021.123046