If you're one of the many chemists who feel that successful crystallization is part science, part talent, part experience, and part witchcraft, I don't know if this paper is going to change your mind or just intensify those feelings. It's addressing a problem that not many people are aware of: recovering lactose from the large amounts of whey produced by the cheese industry. Lactose itself is pretty useful, but although its concentration in various whey liquids can be high, isolating it reproducibly from these is apparently quite a pain.
Having done my PhD in a group that used carbohydrates as chiral starting materials, I can well believe it. We're using to thinking of "sugar" as an incredibly crystalline substance, but that one (sucrose) is abnormally well-behaved when you consider the landscape of mono- and di-saccharides. Most all of them can indeed be isolated as crystalline materials, but the problem is that sometimes each individual sugar can be isolated as several crystalline materials, depending on concentration, temperature, stirring, amount of water present, and so on. So you have that to contend with, and it's also true that some of them just seem reluctant to crystallize at all. Fructose, for example, was known for decades as the "uncrystallizable sugar" - thick gooey syrups all the way, no matter what you tried. As I recall, a large-scale attempt at crystallizing it finally paid off, when fructose solutions of various concentrations were "seeded" with other crystalline carbohydrates, and the one touched off by pentaerythritol finally produced crystalline fructose, which could then be used in turn to seed/nucleate further batches. It's possible, I suppose, that all the crystalline fructose in the world traces back to that experiment somehow, if no one ever found a reliable non-nucleating method to get the stuff. It is true that fructose is largely handled as a syrup even now, mostly as a 55:45 mixture with glucose (the famous/infamous high-fructose corn syrup).
The crystallization of lactose illustrates some of the weirdness of the whole process. In general, crystallization occurs from saturated solutions of some compound - supersaturated, actually, which means that the solvent is carrying more of the solute than it stably can. Some event will eventually make supersaturated solutions "crash out", but the key is getting that to happen in a controlled fashion. Some substances at this point will just "oil out" and make a separate liquid phase, of course, and even normally crystalline ones will do that if there are too many other impurities present. But real crystal formation also happens in this metastable zone of supersaturation, and the interesting/annoying part is that the width of that metastable zone depends on the solvent (naturally), the solute that's going to crystallize out (naturally), and (rather unnaturally!) the sort of vessel that you're doing all this in, its currents and eddys, the type of stirring it's undergoing, the roughness or smoothness of its sides, and who knows what else. Lactose, as fate would have it, tends to have a rather wide metastable zone, which means that it can get pretty heavily supersaturated before it decides to rain down all at once. That's not what you want, because that everybody-out-of-the-pool effect tends to lead to a mass of really fine particles rather than decent crystals, which generally grow a bit (or a lot!) more slowly. These aren't the only two options - in the real world, you can end up with all kinds of particle distributions, from all dusty fine particles to all chunky crystals, with lots of bimodal mixtures in between or worse.
As the article goes into detail explaining, a lot of industrial crystallization takes place in stirred tanks, and one known way to shrink that metastable zone width is to rev up the stirring. But that can be taken too far - powerful stirring will in turn likely break up the nice crystals that initially form and leave you with more of the powdery residue that you were trying to avoid (it causes difficulty in filtration, among other problems). You can also try to slow down the rate that the solution reaches supersaturation, but these mean that each batch in the tank will take much longer, which may not suit your needs, either. An "oscillatory dynamic baffled crystallizer" (ODBC) is an alternative that is supposed to greatly improve mixing without generating a lot of shear forces that break up the crystals (as opposed to the good ol' stirred tank method). It's essentially a more oblong vessel with a cylindrical piston in the middle of it that moves up and down. The paper has several references to head-to-head comparisons of ODBC and STC techniques, and the former does look promising. But lactose has a much nastier reputation for crystallization problems than those literature examples, with a notoriously wide metastable zone under many conditions.
The ODCB comes out looking pretty useful here, too. The authors found that it did indeed reproduce the sorts of results that you'd get at much higher stirring rates in an STC, and what's more, they found that as you went to higher and higher supersaturations that it didn't seem to matter as much how slowly you cooled the solution down. The lower supersaturations were both more sensitive to cooling rate and also tended to produce a wider range of particle sizes, whereas the highly saturated solutions gave a narrower size distribution (which is generally what you'd like to see). On the other hand, the fast cooling technique led to a somewhat lower overall recovery of the starting lactose load, so you'd have to balance the time/throughput advantage versus that problem, and the particle sizes themselves never got to the highest values that could be reached in some stirred-tank conditions. The team also tried another technique (direct nucleation control, or DNC) which takes the batch through repeated heating and cooling cycles. This took some tuning up, because under some conditions the heating would overshoot and bounce around a bit. But under the best conditions, the particle size distribution was narrow (good), the recovery was high (good), but the particle size had somehow gotten even smaller (bad!)
Overall, the ODCB looks promising here, but there are still some things to be worked out. And this sort of study shows why you have to run the experiments, because you can see as you read it that the authors ran into several things that they weren't expecting along the way. There are too many things going on to trust modeling - perhaps as the landscape for a given compound in a given crystallization system gets more worked out you could come up with a good model to predict the interplay between throughput, recovery, particle size, and size distribution, but that would be different for each substance that you're trying to crystallize, of course. As mentioned, though, lactose seems to be a pretty good test for this sort of thing, with a combination of challenging behavior and real-world need for improved methods. If you have a great idea for a new crystallization technique, here's your opportunity. . .