The Test That Takes Them in Turn

I.The First Known Thing — the stages that are not the process

numerical analysis · the splitting of an evolution equation

A great many processes are written as one equation in which several mechanisms act on a single state at the same time. A solute diffuses while it is carried along and while it reacts; each mechanism has its own natural method, and the methods do not combine. The standard recourse is to split the problem: advance the state under the first mechanism alone for a step, then advance the result under the second alone for a step, and call the pair one step of the whole. It is the oldest trick in the subject and it is used because it works. What it produces, however, is not the state the concurrent process would have reached. Taking two mechanisms in turn is a different operation from running them together, and the difference does not vanish merely because each stage was computed exactly.

the quantity that measures the difference, and what it is not a function of

The size of that difference is known in closed form. Expanding the product of the two stage operators against the operator of the concurrent process — the Baker–Campbell–Hausdorff expansion does the bookkeeping — leaves a leading term proportional to the square of the step and to the commutator of the two mechanisms, the amount by which applying them in one order differs from applying them in the other. If the mechanisms commute, the splitting is not merely accurate but exact, at any step, however coarse. The error of a staged computation is therefore a property of the relation between the mechanisms and not of the magnitude, the speed or the importance of either one. Two violent mechanisms that commute cost nothing to stage; two mild ones that do not can cost a great deal.

two levers, and their returns are not the same

From this the discipline draws a practical ordering that is worth stating plainly, because it is the part that travels. There are two ways to reduce the error. The first is to take smaller steps: the sequential arrangement is first-order accurate, so halving the step halves the error. The second is to make each step symmetric — half a step of the first mechanism, a full step of the second, half a step of the first again — whereupon the commutator term cancels against itself and the leading error becomes a double commutator at the square of the step. Halving the step now quarters the error. The symmetric arrangement does the same quantity of arithmetic as the sequential one. The subject’s standing advice is accordingly not to take smaller steps but to take symmetric ones, and the refinement of the step is the weaker of the two levers even though it is the one that looks like diligence.

the limits the field states for itself

The bounds are not unconditional, and the literature is careful about it. They require some regularity in the mechanisms; where one is stiff or unbounded the nominal order can be lost, and the constants hide behind the exponents. Where one mechanism is much faster than the other, the splitting error routinely exceeds the discretisation error of either stage taken alone, which is the practical reason the commutator and not the step size is the first thing to look at. None of this is controversial. It is a settled piece of applied mathematics with a century of use behind it.

II.The Second Known Thing — the test built in two stages

drug discovery · formulation and delivery

A large share of modern oral candidates are poorly soluble weak bases, and for such a compound the fasted stomach and the fasted small intestine are chemically opposite environments. In acid the base is ionised and dissolves freely; at intestinal pH it is not, and a solution carried across the pylorus arrives supersaturated and may stay that way for hours or may precipitate within minutes. Which of those happens decides how much drug is absorbed, and nothing about the molecule settles it, since the outcome depends on the formulation, on the excipients, on the solid form that comes out of solution and on the bile salts and lipids present to hold it up.

the instrument, and the number taken from it

The laboratory instrument for this is the two-stage, or two-tier, dissolution test. The dose is held in an acidic medium for a gastric interval, the vessel is then brought to intestinal pH, commonly by adding a buffer concentrate, and the measurement continues in a buffered, bile-salt-containing medium meant to resemble intestinal fluid. The reading is dissolved concentration against time, and what the formulator carries away from it is not the curve but a parameter: a precipitation rate constant, with a supersaturation at which it was observed, which is then handed to a physiologically based absorption model to predict a plasma profile. For this class of compound that constant is widely acknowledged as the weakest input the model receives.

the objection the field has already made, and the apparatus built to answer it

The field has long held that one abrupt change of medium is not what a dose experiences. The stomach does not convert to an intestine; it empties, gradually, at a rate usually treated as first order with a half-time of some minutes, so that while early material is already supersaturating downstream the remaining solid is still dissolving in acid. The answer has been to gradualise the change: the transfer model, introduced for exactly this class of compound, pumps the acid vessel into the intestinal vessel at a controlled rate, and its descendants add compartments, partitioning phases and absorptive sinks. These predict better than the single abrupt shift, and the objection they answer is specifically about abruptness — about the size of the change made at once.

where the staging survives, and where it does not

It is worth being exact about which half of the discipline stages anything. The simulation half does not. A physiologically based absorption model holds its compartments as a coupled system and advances them together; gastric dissolution and intestinal precipitation proceed concurrently in the solver, as they do in a patient, and the solver can be and is checked compartment by compartment. The apparatus is the part that takes them in turn. So whatever the staging costs does not appear as an error in the prediction. It appears earlier, as a bias in the constant the correct solver is given, and it is invisible to every check performed on the solver.

the run that is never made

One experiment is absent from the catalogue. Nobody runs the intestinal medium first and the acid afterwards, because there is no gut that order resembles. The instrument catalogue of the field is organised by biorelevance, and a reversed run has none.

III.The Bridge

no paper on splitting methods cites a paper on dissolution testing; both halls reproduce a concurrent process by taking its mechanisms in turn, and only one of them knows what that costs or how to get an order back

The halls do not meet. Splitting methods are numerical analysis, their journals are those of computational mathematics, and their examples are advection, reaction and quantum evolution; biopredictive dissolution is pharmaceutics, and its journals are those of formulation science. Neither cites the other and no review spans them. Four neighbours have to be set aside before the proximity is worth anything. The existing objection to the abrupt shift is a statement about step size, which the first hall agrees with and regards as the weaker of its two levers. The multi-compartment apparatus is a refinement of the same kind: adding compartments divides the step more finely and does not symmetrise it. Order effects in dissolution are known for particular compounds, where acid pre-exposure changes the solid form that later precipitates, but that is a chemical mechanism in one molecule and not a quantity governing a test’s error. And the coupled solver, which genuinely does run the mechanisms together, is the reason the problem is easy to overlook: the staging has already happened by the time the solver is handed its inputs.

X: where a concurrent process is reproduced by taking its mechanisms in turn, the discrepancy is first order in the commutator between them, measurable by reversing their order, and removable by an order through symmetry rather than through refinement

Call it commutator-bounded staging. Where a process in which several mechanisms act at once upon one shared state is reproduced by applying those mechanisms in turn, the discrepancy between the staged reproduction and the process is first order in the commutator of the mechanisms and not in the magnitude of either, so it is a property of their relation; being so, it is measurable by the single experiment that applies them in the opposite order, which has no other use and need resemble nothing. Refining the stages reduces the discrepancy to the first power of the refinement, while symmetrising them — half of the first mechanism, all of the second, half of the first again — cancels the leading term and reduces it to the second power, at no additional work. And where the staged procedure is not the simulation of the process but the measurement from which the simulation takes its parameters, the discrepancy is not an error in a prediction but a bias in a fitted constant, carried into a solver that is itself correct and checkable.

three consequences, not one said three times

Three things follow. The first is that the cost of staging has a direct measurement that nobody has an incentive to make: the reversed run calibrates the forward run, and its lack of biorelevance is not a defect in it but the reason it is informative, since what it measures is a property of the apparatus and not of a gut. The second is that fidelity and concurrency are separate purchases, and the field has been buying one of them: more faithful media and more compartments both refine the step, and neither symmetrises it, whereas a partial return of intestinal content to the acid vessel — a second pump, running the other way, on a schedule matched to the forward one — is a symmetrisation and would buy an order where another compartment buys a factor. The third is where the bias lands: not in the prediction, where it would be found, but in a constant inside a coupled solver, which is why a model may be validated compartment by compartment, found sound, and still be wrong by the commutator of the apparatus that fed it.

IV.The Question That Becomes Possible

neither discipline would ask this alone

The numerical analyst has no reason to ask it. His stages are operators he writes down himself, his states are vectors, and he has no dose, no vessel and no fitted constant that something else will later use; the question of what a laboratory instrument is a splitting of does not arise in his hall. The formulator has no reason either, because his design variable is fidelity. His instrument is judged by how closely its media and its transfer rate resemble a fasted human, the commutator is not a quantity in his vocabulary, and the one run that would measure it is excluded by the very criterion that organises his catalogue. Held together, the halls pose a question that is neither’s: whether the error of a staged dissolution test is governed by a commutator one can measure, and whether a return flow nobody has a physiological argument for would buy an order of accuracy.

the commutator-index hypothesis

Concretely. For a given compound and formulation, run two sequences in the same apparatus with the same media, the same hydrodynamics and the same total times: acid then intestinal medium, and intestinal medium then acid. Define the commutator index as the time-integrated absolute difference between the two dissolved-amount trajectories, divided by the integral of their mean, so that it is dimensionless and zero when the order makes no difference. The claim has two parts. First, that this index predicts the size of the discrepancy between the precipitation rate constant fitted from the forward staged run and the constant required to reproduce the observed human plasma profile — a discrepancy the field currently absorbs by refitting. Second, that a symmetrised transfer schedule, with a matched return fraction, reduces that discrepancy by an amount proportional to the index, and leaves it unchanged for compounds whose index is near zero. The second half of that is the internal control, and it is what distinguishes the claim from a general assertion that more mixing helps.

why it would matter

Because the precipitation constant is the weakest number in oral absorption prediction for this class, and because the remedy, if the index behaves, is unusually cheap. It requires no new medium, no new analytical method, no animal and no clinical study: one reversed run per compound to measure the index, and a second pump to symmetrise the schedule. It would also give the field a reason to perform an experiment that its own organising criterion presently forbids, which is a thing worth having independently of how this particular index turns out.

V.How It Could Be Proven — or Broken

a candidate must be breakable, or it is merely poetry

An intercognix candidate earns nothing until it names its own test. This one needs a bench, not a cohort. The compounds, the media and the published human profiles exist; what has to be generated is one extra run per compound and one modified transfer schedule.

the honest status

Note the narrowness. The claim concerns the bias of a parameter fitted in staged apparatus, for compounds whose solubility depends on pH, and it is silent on everything around that. It does not say which media are right, which is the question the field is actually arguing about. It holds permeability and absorption fixed and asserts nothing about either. It does not claim that a return flow is physiologically faithful: the return flow is justified by the arithmetic of symmetrisation, and the fact that the gastroenterological literature has a name for a retrograde duodenal flow is at most a reason the device is not absurd, not an argument for it. And where solubility does not depend on pH the index is zero by construction and the proposal is empty, which is the correct behaviour for it and not a saving qualification.