Full opinion text
OPINION RICHARD A. POSNER, Circuit Judge. I conducted a bench trial in this patent infringement case between February 5 and 21 of this year, by designation of the chief circuit judge' pursuant to 28 U.S.C. § 291(b), and this opinion comprises the findings of fact and conclusions of law that Fed.R.Civ.P. 52(a) requires me to issue. The plaintiffs in their post-trial brief erroneously refer to “findings” that I made in the course of the trial. I made no findings during the trial; these are my findings. The Procedural Setting and Pretrial Rulings The plaintiffs are two subsidiaries of Glaxo, the British manufacturer of pharmaceutical drugs. I shall refer to the two collectively as “SmithKline”. The defendants are Apotex, Inc., a Canadian manufacturer of generic pharmaceuticals, and affiliates of it; I shall refer to the defendants collectively as “Apotex.” Smith-Kline claims that Apotex is (more precisely, as I’ll explain in due course, will be) infringing SmithKline’s U.S. patent 4,721,-723, or “723” as the parties refer to it, after its last three digits. The patent is on an antidepressant drug that SmithKline sells under the trade name Paxil. Paxil is now the leading brand of antidepressant drug, with annual sales of $3.2 billion worldwide, about two-thirds in the United States. SmithKline argues not only that Apotex is an infringer but also that the infringement is willful, entitling SmithKline to additional relief, namely an award of attorneys’ fees. Apotex contends that the patent (more precisely, the claim in the patent that covers the composition of the drug, as distinct from claims concerning processes for making the drug) is invalid but, if valid, not infringed, and if valid and infringed still not infringed willfully. I deferred consideration of the issue of willfulness to such time as I might hold the patent valid and infringed. Since I hold that the patent although valid is not infringed and further that even if it is valid and infringed SmithKline is entitled to no relief, I do not reach the issue of willfulness. Should the case be remanded to me, I would have to conduct a further evidentiary hearing in order to resolve that issue. Judge (now Chief Judge) Kocor-as of the Northern District of Illinois, to whom the case was initially assigned when it was filed back in 1998, made a number of pretrial rulings, and I must consider the extent to which I am bound by them in accordance with the doctrine of law of the case. The doctrine requires a court to adhere to its previous rulings in the same litigation unless there is a compelling reason, such as an intervening change of law or newly discovered evidence, to reexamine them. Its usual application is to a case that is remanded by the court of appeals and then returns to that court by an appeal from the decision on remand. The doctrine counsels the court not to revisit the issues it decided on the first appeal, but it does not limit a trial judge’s changes of mind during the course of a litigation uninterrupted by an appeal, and such changes of mind are of course frequent. When however the judges in a case are switched in midstream, as happened here, the successor judge may not reconsider his predecessor’s rulings with the same freedom that he may reconsider his own rulings. “As this court [the Seventh Circuit] explained in Williams v. C.I.R., 1 F.3d 502 (7th Cir.1993), the law of the case doctrine in these circumstances reflects the rightful expectation of litigants that a change of judges midway through a case will not mean going back to square one. See also Christianson v. Colt Industries Operating Corp., 486 U.S. 800, 816-17, 108 S.Ct. 2166, 100 L.Ed.2d 811 (1988). Although the second judge may alter previous rulings if he is convinced they are incorrect, ‘he is not free to do so ... merely because he has a different view of the law or facts from the first judge.’ Williams, 1 F.3d at 503. Instead, the presumption is that earlier rulings will stand, even though it can be overcome for compelling reasons (such as new controlling law or clear error).” Best v. Shell Oil Co., 107 F.3d 544, 546 (7th Cir.1997); see also Peterson v. Lindner, 765 F.2d 698, 704 (7th Cir.1985); Fagan v. City of Vineland, 22 F.3d 1283, 1290 (3d Cir.1994). The reader may wonder at my citing Seventh Circuit rather than Federal Circuit cases on this point, since if my decision is appealed it will be appealed to the Federal Circuit. But that court applies the law of the circuit in which the district court is located to procedural matters that are not unique either to patent law or to appellate as distinct from trial procedure. See Panduit Corp. v. All States Plastic Mfg. Co., 744 F.2d 1564, 1574-75 (Fed.Cir.1984) (per curiam); Biodex Corp. v. Loredan Biomedical, Inc., 946 F.2d 850, 858 (Fed.Cir.1991). The doctrine of law of the case is not unique in either sense and so its application here is governed by Seventh Circuit (and of course Supreme Court) precedent. The parties asked me to reexamine several of Judge Kocoras’s rulings, but only one met the criteria for reexamination, his order of July 16, 2002. That order excluded evidence that infringement of SmithKline’s patent 723 would occur when Apotex’s product combined with the fluids in a patient’s stomach to create small amounts of the patented product, or perhaps even earlier when the patient opened the bottle of tablets and took out a pill, reexposing it as well as the remaining pills in the bottle to air and hence to moisture. The patient would be the in-eases (assuming that there is any infringer in those cases, which I express no view on), but Smith-Kline argues that Apotex, knowing that infringement would occur, would be guilty of inducement to infringe. 35 U.S.C. § 271(b). As it pointed out in its motion to vacate the July 16 order, the Federal Circuit’s decision in Warner-Lambert Co. v. Apotex Corp., 316 F.3d 1348 (Fed.Cir.2003), undercuts the ground of that order and so reexamination is warranted. But Warner-Lambert also makes clear that the burden of proving inducement is a heavy one, id. at 1363-64, and Smith-Kline’s motion provided no ground for thinking that SmithKline could carry the burden. Even if a patient’s gastrointestinal juices convert the nonpatented product that Apotex plans to manufacture to the product patented by SmithKline and Apo-tex knows this will happen, there is no evidence that Apotex intends, in the sense of desires or is working to achieve, this result. For the gastrointestinal “infringement” does nothing for Apotex commercially; it merely increases Apotex’s exposure to liability. That is equally true if infringement is thought to occur when the bottle is opened. Apotex has tried to prevent conversion of its product to the patented form and a principal issue in this case is whether it has succeeded; there is suggestion that Apotex desires conversion. I therefore denied the motion to vacate the July 16 order and I adhere to that ruling. But I add that, in light of the discussion of relief in the last part of this opinion, it is plain that if Apotex were guilty of inducement in the circumstances outlined above, SmithKline would not be entitled to any relief. The Background of the Case Early in 1977 a British company called Ferrosan obtained a U.S. patent (“196,” also known as the Christensen patent) on a set of compounds that included what came to be called “paroxetine.” The pa-roxetine molecule consists of carbon, nitrogen, oxygen, hydrogen, and fluorine atoms arranged in a particular configuration. When combined with additional atoms to form a salt molecule (a hydrochloride, for example, if paroxetine base is bathed in hydrochloric acid), and mixed with additional compounds (binders, lubricants, disintegrants, etc.-eollectively, “excipients”) to bulk up into a pill and to improve handling, tableting, and dissolution, paroxe-tine was believed, correctly as it turned out, to be effective in treating depression and related psychiatric disorders. Like fluoxetine-the active ingredient in Prozac-paroxetine is a selective serotonin reup-take inhibitor, which helps to assure an adequate supply of the “feel good” enzyme serotinin to brain cells. Although mention of Prozac is a reminder that Paxil faces competition from other SSRIs (not to mention other types of antidepressant drug, and other modes of treatment altogether), there are medically significant differences, both in efficacy and side effects, even among the different SSRIs. Paxil, for example, is preferred to Prozac by many doctors for the treatment of depression coupled with anxiety (a common combination), because unlike Prozac it has been approved by the Food and Drug Administration for anxiety disorders. Ferrosan was not a manufacturer of pharmaceutical drugs, so in 1980 it licensed its paroxetine patent to Smith-Kline. Although the patent specified pa-roxetine maleate as the paroxetine salt it was claiming, Ferrosan had already, after some travail, succeeded in creating a pa-roxetine hydrochloride in crystalline form, hydrochloride being a preferred salt for pharmaceutical purposes. In 1981 Smith-Kline began manufacturing paroxetine hydrochloride in its Harlow (U.K.) plant. Before a pharmaceutical drug can be placed on the market, it must undergo elaborate testing for safety and efficacy, and so quantities of paroxetine hydrochloride were distributed to different parts of the world, including the United States, for use in clinical trials. SmithKline has a laboratory in Worthing, England, and samples of paroxetine hydrochloride were sent there in bulk form-that is, before being made into pills-to be used in experiments on improving the production of the bulk material. In March 1985, a chemist at Worthing, Alan Curzons, experimenting on ways to produce paroxetine hydrochloride, discovered that he had created a new form of the compound, which he dubbed “Form 1” to distinguish it from the anhydrous form, which he dubbed (confusingly, because it was the earlier form) Form 2. Tests that Curzons performed confirmed that Form 1 was indeed a distinct crystalline form of paroxetine hydrochloride. Crystallinity is central to this case. When molecules are bound together, by interatomic forces that radiate beyond the “boundaries” of the molecules themselves, in a definite structure which is then repeated over and over again without significant change, the agglomerations that result are called “crystals.” (These and other relevant aspects of crystallography are lucidly discussed in Stephen R. Byrn, Ralph R. Pfeiffer & Joseph G. Stowell, Solid-State Chemistry of Drugs (2d ed.1999). See especially Chapters 1, 10-11, and 13. Dr. Byrn testified for Smith-Kline at the trial.) The molecules are like the intersections of the slats of a lattice; the slats correspond to the forces that hold the molecules in their fixed positions, and the multiplication of the lattice is the crystal. The crystal’s minimum structure-the structure that, repeated, constitutes the crystal-is called the “unit crystal cell.” The unit crystal cell is not itself a crystal, however. That is by definition: a crystal is a multiple of the structure that defines the unit crystal cell. Moreover, were there only, say, two unit crystal cells, the molecules composing them would not crystallize because the intera-tomic forces would be too weak to maintain the structure. The number of unit crystal cells required to create the minimum crystal is very small, however: depending on the molecules and the structure, as few as ten molecules may be enough to create an actual crystal. Stated differently, a large crystal might in principle though not in practice be cut into an immense number of utterly minute crystals. The same substance will sometimes appear in more than one crystalline form-will be, that is, “polymorphous.” The molecules are the same but the lattice is different. The difference can affect the melting point of the crystal (the point at which the crystal structure is destroyed by heat, rendering the substance liquid) and other properties of the crystal, such as hardness: a dramatic example is graphite and diamonds, both of which are crystals of carbon. Because a different arrangement of molecules implies a different pattern of bonds, and different bonds vibrate at different frequencies, different polymorphs of the same chemical produce different x-ray diffraction patterns and infrared spectra, which are two types of graphic mapping of the atomic forces binding the crystal. The form of paroxetine hydrochloride that Curzons discovered was not a true polymorph, although often and loosely referred to as such, as I shall do; rather, it was a “pseudopolymorph.” These critters not only have their molecules arranged differently but also have a slightly different molecular composition. A common type of pseudopolymorph is a solvate, which is a crystal in which molecules of a solvent, such as water, have become “caught” inside the crystal and have bonded with the other molecules in an altered crystalline structure. When the trapped and bonded solvent is water, the solvate is called a hydrate. And when it is a hydrate in which there is one water molecule for every two of the other molecules constituting the unit crystal cell, in this case molecules of paroxetine hydrochloride, the hydrate is called a hemihydrate. Despite the presence of water molecules, a hemihydrate is a solid, a powder, at room temperature. At the time that Curzons had his eureka moment, SmithKline’s plant at Harlow was producing only a paroxetine hydrochloride anhydrate (at least as far as it knew), which is to say a crystalline form of parox-etine that does not contain a bound water molecule. (This anhydrate is what Cur-zons dubbed “Form 2.”) Curzons made a batch of paroxetine, added isopropyl alcohol, a solvent, and found that the batch crystallized as a hemihydrate rather than as an anhydrate. And here an oddity, as it strikes a lay observer at any rate, should be noted. The anhydrous form of crystalline paroxetine hydrochloride is hygroscopic; that is, it attracts water, perhaps because of the position of the fluorine atoms in the anhydrous form, though this was merely a speculation by one of the expert witnesses. The water it attracts either sits on the outside of the paroxetine molecule, loosely attached and therefore easily dispersed by heating at a significantly lower temperature than required to liberate the bound water molecule from the hemih-ydrate, or, if it has found its way inside the crystal, it is nevertheless again readily dispersed, because it is not held to the parox-etine molecules by strong interatomic bonds. In contrast, the hemihydrate is not hygroscopic because it is not “thirsty”-it has already drunk, as it were. The anhydrate’s hygroseopicity makes it difficult to handle in the manufacturing process; measures must be taken to control humidity and other sources of moisture lest the anhydrate become so “soggy” that it degrades into other compounds, which might impair the safety or efficacy of the product. So when Curzons realized that he had obtained a hemihydrate he immediately grasped the potential commercial significance-and also and distinctly the potential patent significance, which has now to be explained-of his discovery. Because it takes a long time for a new drug to be approved by the U.S. Food and Drug Administration for sale to the American public, the actual period during which the producer has an exclusive right to make, use, and sell the drug is shorter than the statutory term of the patent. In the ease of patent 723 for example, the patent at issue in this case, the application was filed in 1985 and granted in 1988, and the patent expires in 2006; but because the FDA process delayed the commencement of commercial sale to 1993 (FDA approval having been obtained the previous year), the effective term of the patent has been compressed to 13 years. Indeed, were it not for 35 U.S.C. § 156(c), a provision added to the patent statute by the Hatch-Waxman Act (the Drug Price Competition and Patent Term Restoration Act of 1984, Pub.L. No. 98-417, 98 Stat. 1585 (1984)), SmithKline would have had only 12 years of effective patent protection, because, were it not for that provision, patent 723 would expire in 2005-20 years after the date of application and 17 years after the date the patent was issued-rather than in 2006. The compression of the commercially significant patent term by reason of the regulatory process at the FDA is a matter of concern to the manufacturers of new drugs. The cost of developing such a drug is often very great, in part because attempts to develop a new drug that will be both safe and effective often fail and the cost of these “dry holes” must be reckoned into the cost of the drugs that succeed, as it is only out of the revenues of those drugs that the costs of the dry holes can be recovered. The greater the upfront cost of developing a product, the more time is required to recoup the cost and so (other things being equal) the longer is the socially optimal patent term. The costs incurred in running the gauntlet of FDA approval not only increase the manufacturer’s upfront development cost but compound the effect of the delay, also due largely to the FDA, between obtaining a patent and actually being able to market the patented drug to the consuming public. Although Ferrosan obtained its patent on paroxetine in 1977, the product still had not come to market as of 1985, when it was superseded in SmithKline’s patent planning by the hemihydrate discovered that year. Until the Hatch-Waxman Act was passed in 1984, the costs and delays imposed by the FDA’s procedures worked in favor of as well as against the manufacturers of patentable drugs. The reason is that generic manufacturers (such as Apotex), that is, manufacturers of drugs that have come off patent, either because the patent has been invalidated or, more commonly, because the patent has expired, also have to obtain the FDA’s approval before they can sell the generic drug in the United States. And, before Hatch-Wax-man, the generic manufacturer could not, in demonstrating to the FDA that his generic version would be no less safe and effective than the patented original, rely on the animal and human tests conducted by the manufacturer of the patented drug. He had to do his own tests. This required him to make or use the patented product, but he could not do so lawfully before the drug came off patent unless he had a license from the patentee-otherwise he would be an infringer because the Federal Circuit had held that the “experimental use” defense to patent infringement was inapplicable to experiments having commercial aims. Roche Products, Inc. v. Rolar Pharmaceutical Co., 733 F.2d 858, 863 (Fed.Cir.1984). The upshot was that the generic manufacturer could not begin the process of seeking FDA approval until the patent expired, and given the length of that process the practical effect was to extend the period of patent protection well beyond the statutory term. The Hatch-Waxman Act shortened the process both by allowing the generic companies to take a free ride on the results of the patentee’s safety and efficacy testing so long as they could show that their product was bioequi-valent to the original and by allowing them to make and use the patented product, even though the patent hadn’t yet expired, in order to demonstrate bioequivalence. This case differs from the standard case to which Hatch-Waxman applies because Apotex claims to be making a drug that while bioequivalent to a patented drug does not infringe the patent because it is a different compound. Apotex still had to convince the FDA of this bioequivalence, and whether to aid in demonstrating this or (more likely) merely to learn more about the production of paroxetine, Apotex bought some Paxil tablets, extracted the hemihydrate from them, and even made its own hemihydrate. Although this experimentation amounted to a making or use (in fact both) of the patented hemihydrate, SmithKline concedes that it was not infringement because it fell within the expanded experimental-use privilege created by the Hatch-Waxman Act. The Act allows the generic manufacturer to file an Abbreviated New Drug Application (ANDA)-abbreviated because the applicant has only to demonstrate bioequiva-lence, and not safety and efficacy as an original matter-before the patent expires on the drug that it wants to produce and sell, though it may not begin to sell the drug until the patent either expires or is declared invalid. If the manufacturer wants to sell its generic equivalent before expiration of the patent, on the ground that the patent either is invalid or won’t be infringed because while the two drugs are bioequivalent they are not the same invention, he has to so state in his ANDA. The Act authorizes the patentee in such a case, if he doesn’t agree that the patent is invalid or even if valid will not be infringed, to sue the generic manufacturer for infringement, even though there is no infringement in the usual sense because the generic manufacturer has not yet begun to make (other than for permitted experimental purposes) or sell his knock-off of the patented drug. Had the Act stopped there, its unequivocal effect would have been to shorten the economically significant patent term of drugs. But as a concession to the manufacturers of patented drugs, who had complained with much support in the academic literature about the length of time it took to get a new drug approved by the FDA, Congress tolled the date of expiration of drug patents (the patent term for all utility patents, including therefore drug patents, was then 17 years but in 1995 it was increased to 20 years) during the period in which an application for a new drug was under regulatory review. The Act capped the extension for patents such as 723 issued after the Act • was passed at five years. 35 U.S.C. § 156(g)(6)(A). And patent protection could not extend beyond 14 years after the FDA had approved the new-drug application. 35 U.S.C. § 156(c)(3). The manufacturers of patented drugs were not happy about the trade, in part at least for a reason that the facts of the present case illuminate. The Ferrosan patent (patent 196) expired in 1992. Yet as I have pointed out, as late as 1984, with only eight years to expiration, paroxetine made under that patent had not yet been placed on the market. If, however, the hemihydrate version of the molecule could be patented, the effect might be a considerable extension in the effective patent term of paroxetine because it might become difficult or even impossible to manufacture the pure anhydrous form after the Ferrosan patent expired. How so? Dr. Joel Bernstein, one of SmithKline’s expert witnesses at the trial, is an authority on “disappearing poly-morphs.” See Joel Bernstein, Polymorphism in Molecular Crystals 89-92 (2002); J.D. Dunitz & J. Bernstein, “Disappearing Polymorphs,” 28 Accounts of Chem. Res. 193 (1995); see also Byrn, Pfeiffer & Stowell, supra, at 463. The term refers to the fact that after a new polymorph or pseudo-polymorph appears, the process that had been used to make the old polymorph may no longer produce it-may produce instead the new one. Actually there’s an ambiguity buried in this formulation that is important to this case. A polymorph could disappear in the literal sense that it could no longer be created; or it could disappear in the more limited sense that the pure form of the old polymorph could no longer be created-the new polymorph would be an indelible though possibly minor and functionally inert component of any batch of the old. The causal mechanism of polymorphic creation and transformation is not dear-polymorphism simply is not well understood by science, because of the complexity of the atomic interactions that produce it-but it seems to involve the confluence of three factors. First, according to Ostwald’s Rule, a well-attested scientific principle, later-appearing polymorphs tend to be more stable than the earlier ones. If there is metamorphosis (interconversion, chemists call it), it tends naturally to be in the direction of more stable forms. A more stable form is less likely to change into a less stable one than vice versa, and so there is a drift toward the more stable polymorphs. Second, impurities retard crystallization, including crystallization in new forms, and the progress of technology has yielded a secular decline in the proportion of impurities in manufactured products. And third, once a new and more stable crystal emerges, should it be mixed, even in very small quantities, with the old, less stable crystal, the old form may convert to the new. This process of “seeding” the old with the new can be deliberate-that is, can be a method of manufacturing the new polymorph-or adventitious, a result of the fact that some of the crystals become airborne and “contaminate” the laboratory or plant in which the old crystal is being manufactured. I must pause over the terms “seed” and “seeding” because of the importance they assumed in the trial. In its broadest crystallographic sense a seed is any bit of matter that precipitates crystallization; it could be a grain of dust. But the seeds relevant to this case are seeds that cause one polymorph to convert to another and these seeds are crystals of the form to which conversion occurs. See Bernstein, supra, at 90-91. A single tiny crystal, constituting a single seed, might induce conversion. Id. at 91. The first two factors that I have discussed under the rubric of “disappearing polymorphs” together provide the most persuasive explanation for the initial appearance of a new polymorph, while the first and third provide the most persuasive explanation for the “disappearing poly-morphs” phenomenon itself. The creation of the new polymorph is likely to make the laboratory or plant where it is produced seeded, with the result that efforts to produce the old polymorph may instead produce the new one, since it is the more stable form. In principle it should be possible to re-create the old polymorph, just by repheating the exact procedure by which it used to be created, only this time in a seed-free environment. Although it is difficult, and in some cases it may be impossible (paroxetine hydrochloride hemih-ydrate may be one of those cases-no one knows), to destroy all the seeds in seeded premises, crystalline seeds, unlike the pods in Invasion of the Body Snatchers, do not traverse galactic distances under their own power. Unless they are carried in samples or on a person’s clothing from one seeded premises to another, the new premises will not be seeded and so it should be possible to re-create the old polymorph there. But this, as Dr. Bernstein explained, is in principle; and in practice efforts to re-create old polymorphs do not always succeed, probably because the critical mass of molecules that is required to cause conversion is so minute. In his book, Dr. Bernstein suggests that “a few tens of molecules” may be the minimum for conversion. Bernstein, supra, at 91. He was writing about polymorphs in general rather than about paroxetine, but it was implicit in Bernstein’s testimony and also that of Dr. Byrn, and not persuasively countered by Apotex’s expert witnesses, that the critical mass of paroxetine hydrochloride hemihydrate crystals required to induce conversion in a batch of anhydrous paroxetine is probably of the same order of magnitude (or minitude). Even if a seed required not ten but ten million molecules, a particle at the limit of visual detection would contain enough paroxetine hydrochloride hemihydrate to make more than one hundred million seeds. And there is no method of ventilation or fumigation that will eliminate all the hemihyd-rate crystals from a manufacturing environment. Dr. Bernstein testified that if Apotex, desperate to avoid a charge of infringement, built a new plant in Antarctica where no hemihydrate seeds had ever been and started manufacturing anhydrate there, and a depressed worker in the plant dropped a Paxil on the floor, the result might be to seed the plant and make it impossible from then on to produce pure anhydrate there. For that matter, he might have dropped it on the floor of his bathroom at home, releasing crystals that adhered to his skin or clothing. Bernstein described a remarkable episode involving the AIDS drug ritonavir, another polymorphic crystal. See also Sanjay R. Chemburkar el al., “Dealing with the Impact of Ritonavir Polymorphs on the Late Stages of Bulk Drug Process Development,” 4 Organic Process Res. & Dev. 413 (2000). Abbott began commercial production of the drug in 1996. Two years later a previously unknown-and, characteristically, a more stable-polymorph (Form II) appeared in the plant in the United States in which the final product was being manufactured. Immediately the old polymorph (Form I) began converting to the new. This precipitated a “market crisis,” id. at 413, because unlike anhydrous and hemih-ydrous paroxetine crystals, Form I and Form II ritonavir were not bioequivalent. Fortunately (or so it seemed), Form II (the new polymorph) had not yet been observed in the plant in Italy where the bulk ritonavir was produced-but shortly after a visit to that plant by scientists who had been exposed to Form II, Form II showed up there too, probably (Bernstein omitted the qualification) as a result of seeding from Form II crystals on the scientists’ clothing. Eventually Abbott was able to produce a version of Form I that would not convert-entirely; but the new version did contain up to 3 percent of Form II. Id. at 417. Because the mechanism of seeding-the process, occurring at the atomic level, by which contact between the more and the less stable polymorph causes the latter to convert to the former-has not yet been discovered, Apotex argues that there is no scientific basis for believing that seeding occurs. But this is obviously wrong. Many scientific phenomena are identified before their causal mechanism is understood. Newton was distressed that he could not identify the causal mechanism behind the law of universal gravitation, which he had discovered, because according to that law bodies at a distance, with no intermediate links, were exerting force on each other. Seeding appears to have been at work in Alan Curzons’ laboratory. One might have supposed that the unexpected result of his experiment had been produced by the exact combination of steps that he had taken in the making and crystallizing of paroxetine, including the choice of solvent to precipitate crystallization. But no; he soon discovered that it was very easy, using a variety of solvents, to produce hemihydrate-and the likeliest explanation is that the first batch of hemihydrate that he created had seeded his lab. This leaves unexplained, however, why his initial experiment resulted in hemihyd-rate. Remember that he was not trying to produce a hemihydrate; the result of the experiment was serendipitous. One possibility is that as a result of the progress of technology, whereby as I noted, over time the proportion of impurities in manufactured products diminishes, the chemical mixture on which he performed his experiment was so far free from impurities that the same application of solvent that had previously produced only anhydrous crystals now produced hemihydrous ones. Another possibility is that batches of paroxe-tine shipped from Harlow to Worthing had already seeded Curzons’ lab because the manufacturing process at Harlow had, though without anyone’s knowing it, already produced hemihydrate. For, once alerted by the outcome of Curzons’ experiment to the existence of the hemihydrate, SmithKline discovered that a batch of pa-roxetine produced at Harlow in December 1984 (HP 28) and another batch produced there in the following month (HP 24) had both been hemihydrate. As SmithKline states in its post-trial brief, “no one knows for sure” how these early batches (including Curzons’) “came to be crystalline pa-roxetine hydrochloride hemihydrate.” Nevertheless Dr. Bernstein testified that he was “absolutely convinced” that no hemihydrate had existed before December 1984. At first glance his testimony appears to lack any scientific basis. Because paroxetine does not exist in nature but, so far as anyone knows, was created for the first time in the early 1970s by Ferrosan, the hemihydrous form could not have existed before then. But it could have come into existence at any time between then and December 1984. It was not detected until March 1985, but existence and detection are not the same thing, for we know that HP 28 and HP 24, which are conceded to be hemihydrate, predated the earliest detection of hemihydrate, which was by Curzons in March 1985. The methods available as late as 1985 did not enable detection of small quantities of the hemih-ydrate in a mixture-smaller than 5 percent at the most. Smaller quantities may have existed ever since the first paroxetine was created. Bernstein thought not, however, and gave two reasons. First, a batch of anhyd-rate manufactured by Ferrosan in 1980, though stored in a hot and humid place (the greater the heat-short of the melting point, of course-and the humidity, the likelier is conversion from the anhydrous to the hemihydrous form), had three years later still not converted to the hemihydrate form, suggesting that it had not been seeded and hence that there were no seeds as late as 1980. And HP 22, manufactured just weeks before HP 23, contained no detectable hemihydrate, whereas HP 23 was entirely hemihydrate. This is evidence that hemihydrate seeds can “metastasize” at a high rate when they come into contact with anhydrate crystals. If so, the fact that HP 23 was the first paroxetine in which the hemihydrate was detected is evidence that there were no seeds before then, that is, before December 1984, and almost certainly not when Ferrosan obtained its patent in 1977. It is trae that Curzons in a 1985 memo reported having discovered hemihydrate in a Ferrosan batch dating from 1980, but quite apart from the possibility of later seeding, Cur-zons convincingly explained that he had been mistaken. Dr. Terence Threlfall, Apotex’s expert on polymorphism, testified to the contrary of Bernstein that anhydrous and hemihyd-rous forms of paroxetine can coexist happily. There is support for this conjecture in SmithKline’s own evidence, of which more later, that some of Apotex’s anhydrous product contains small amounts of hemih-ydrate without conversion of the rest. In other words, as Threlfall testified, a mixture of anhydrate and hemihydrate can be an equilibrium, in which event the earliest batches of paroxetine manufactured by Ferrosan may have contained undetectable quantities of the hemihydrate. In light of this evidence, Dr. Bernstein’s absolute certainty that hemihydrate did not exist before December 1984 is not tenable. No one knows when the hemihydrate form of paroxetine came into existence, although it is a reasonable inference that it did not exist in a detectable amount until then. The conflicting testimony of Bernstein (and also Byrn) on the one hand and of Threlfall on the other can largely be reconciled on the following hypothesis: while the presence of hemihydrate seeds in a batch of anhydrate is likely, provided the ambient humidity and temperature are no lower than is normal in the temperate zone, to produce conversion within a short time, once the amount converted reaches a few percent of the mixture further conversion is unlikely without substantially greater humidity, temperature, or pressure. In other words, if conversion is plotted against time, then in the case of paroxetine hydrochloride as in the case of the revised process for producing Form I ritonavir, one will observe rapid growth from an initial very low level, followed by a leveling off at a few percentage points. This of course is under controlled environmental conditions; given enough humidity, heat, etc., conversion may continue until it reaches 100 percent. By the same token, with much tighter controls less, maybe no, conversion will take place despite the presence of seeds; the clearest case is where there are no water molecules in the environment of the anhydrate. The Patent and the Patent Controversy Crystalline paroxetine hydrochloride hemihydrate, along with certain processes for making it (none challenged by Apotex as invalid or contended by SmithKline to be infringed), was patented in patent 723 in 1988, the patent application having been filed in October 1986. After running the FDA gauntlet, the product was placed on the market under the name Paxil in 1993. The patent will expire, as I noted, at the end of 2006. In 1998 Apotex filed an ANDA in which it proposed to manufacture an anhydrous crystalline form of the paroxetine hydrochloride crystal, patent 196 having expired in 1992. Several other generic manufacturers have since filed ANDAs for various anhydrous forms of paroxetine. Apotex was eager to be first because the generic manufacturer whose ANDA is the first to be approved obtains a 180-day period of exclusivity, see 21 U.S.C. § 355(j)(4)(B)(iv)-180 days during which he and the patentee (if as here the patent on the bioequivalent original has not expired) are in effect duopolists. SmithKline deplores Apotex’s eagerness to be first but of course that eagerness is the mirror of SmithKline’s zeal to retain the much more extensive protection conferred by a patent that excludes the competition of a bioequivalent drug as infringing. The reason Apotex waited as long as it did after the expiration of Ferrosan’s patent to file its ANDA is that the FDA forbids submission of an ANDA on a “new chemical entity,” which crystalline paroxe-tine hydrochloride hemihydrate is, or its bioequivalent (i.e., the anhydrate), until five years after the patented drug has been put on the market. 21 U.S.C. § 355(c)(3)(D)(ii). The hemihydrate wasn’t really that new, being bioequivalent to its predecessor, the paroxetine patented by Ferrosan, but it was deemed new because the FDA had never approved the predecessor. See also 21 C.F.R. § 314.108. As required, the ANDA specified the process of manufacture that Apotex would use to make the product in commercial quantities, and the site, actually sites, of manufacture. The bulk material (the crystalline paroxetine hydrochloride anhydrate before being mixed with excipients and made into pills) would be manufactured in a plant owned by an affiliate of Apotex named Brantford Chemicals, Inc. (BCI). It would then be shipped to a plant owned by another Apotex affiliate, Tor-Pharm, where excipients would be added and mixed, the mixture compressed into pills, and the pills sprayed with an aqueous (88 percent water) coating and bottled for sale. The ANDA represented that Apotex’s anhydrous version of crystalline paroxetine hydrochloride would not infringe patent 723. Disagreeing, SmithKline brought this suit, while meanwhile scurrying to obtain patents on other anhydrous poly-morphs of the paroxetine hydrochloride crystal; these patents are involved in other litigation between it and Apotex. Since 1998, when Apotex filed its ANDA, Smith-Kline or its affiliates have applied for almost 100 patents on paroxetine, though not all, or even most, are on anhydrous forms of the molecule. It may seem odd that SmithKline could obtain any patents on anhydrous paroxe-tine hydrochloride given the expiration of the Ferrosan patent. However, that patent did not refer to paroxetine hydrochloride or to crystallinity, but to a set of compounds of which paroxetine maleate (another paroxetine salt) was one example; and the maleate probably was in amorphous (noncrystalline) form. So there may be room for patents on other salts or other crystals of anhydrous paroxetine. But that is not an issue in this case and I intimate no view on it. Whether manufacturers of brand-name drugs are using follow-on patents to compete unfairly with the generic manufacturers is at' present under investigation by the Federal Trade-Commission (described in an amicus brief that the FTC filed on January 28, 2003, in Smithkline Beecham Corp. v. Apotex Corp., No. 99-CV-4304 (E.D.Pa.)), but played no role in the trial of this case. In pretrial discovery SmithKline obtained samples of crystalline paroxetine hydrochloride anhydrate manufactured by Apotex and submitted them for testing to two academic scientists who testified for SmithKline, Dr. David Batchelder and Dr. Thomas Niemczyk. They testified that they had found hemihydrate in the samples and on the basis of this and other evidence SmithKline contends that the manufacture of the anhydrous product by Apotex is likely to infringe patent 723 because some of Apotex’s output will consist of hemihydrate. According to SmithKline, the BCI plant is seeded with hemihydrate crystals because it was there that Apotex, exercising the broadened experimental-use privilege conferred by the Hatch-Waxman Act, used and made hemihydrate in the course of developing its anhydrous product. Any seed-bearing bulk material produced by BCI will, moreover, be formed into pills by TorPharm; and SmithKline argues that the compression exerted on the bulk material to make pills, together with the temperature and humidity in the plant and the fact that the final stage of manufacture involves the application of a water-based coating, will promote further conversion of the anhydrous to the hemih-ydrous form of paroxetine hydrochloride. Although crystals of paroxetine can be seen by means of an electron microscope if the crystals are “trapped” on the microscope’s stage, there is no technique for detecting minute crystals in particulate matter in a pharmaceutical plant, on a worker’s clothing, etc. But it does not follow, as Apotex unguardedly suggests, that the presence of such crystals, the “seed” crystals, cannot be inferred. What is true and important, however, is that the presence of slight amounts of hemihydrate in Apotex’s product would not “do anything” for Apotex. The anhydrate has the identical therapeutic properties as the hemihydrate; that is entañed by their bioequivalence, which SmithKline does not contest. And although hemihydrate handles better in the manufacturing process because it is not hygroscopic, this wül confer no cost savings on Apotex in the quantities in which it wñl be present (if at all) in Apotex’s product. Dr. Joseph Robinson, an expert on pharmaceutical manufacturing, testified without contradiction that the percentage of hemihydrate in an anhyd-rate mixture would have to be in the “high double digits” to contribute any commercial value to the mixture. From Apotex’s standpoint, any hemihydrate that happens to turn up in its product is a useless byproduct of a manufacturing process that, consistent with the “disappearing poly-morphs” theory, is difficult or perhaps even impossible to conduct without producing tiny amounts of the unwanted byproduct. Dr. Bernstein testified that Apotex was indeed trying to manufacture an anhydrous product, while another of Smith-Kline’s expert witnesses, Dr. Batchelder, testified that hemihydrate in an anhydrous product is an “impurity” and SmithKline’s lead counsel described anhydrate that contains some hemihydrate as “degraded.” Validity and Claim Construction There is no infringing an invalid patent. A number of Apotex’s challenges to the validity of patent 723 were rejected by Chief Judge Kocoras in pretrial rulings that I am not disposed to reexamine. What is left is, first, a frivolous argument that either patent 196, or an early article that discussed the paroxetine molecule, R. Battegay, M. Hager & U. Rauchfleisch, “Double-Blind Comparative Study of Pa-roxetine and Amitryptyline in Depressed Patients of a University Psychiatric Outpatient Clinic (Pilot Study),” 13 Neuropsychobio. 31 (1985), or both, “literally” (as distinct from “inherently,” discussed next) covered the hemihydrate form. There is no mention of hemihydrate in the patent or in the article and the hemihydrate poly-morph of paroxetine was unknown either when the patent application was filed or in 1977 when the patent was granted. Nor has Apotex shown that hemihydrate, the invention claimed by patent 723, was obvious given the prior art (patent 196 and the Battegay et al. article). A more plausible argument is that the hemihydrate form was inherent in patent 196 because anyone who followed the directions in that patent would inevitably produce hemihydrate. If it was inherent, it is deemed anticipated, and patent 723 would be invalid. Atlas Powder Co. v. Ireco, Inc., 190 F.3d 1342, 1347 (Fed.Cir.1999); In re Robertson, supra, 169 F.3d at 745. Although it was not until seven years after patent 196 was granted that Smith-Kline chemists at Harlow, and Alan Cur-zons at Worthing, while following more or less the directions in the patent, produced hemihydrate, the hemihydrous form of pa-roxetine may have existed earlier and if so the fact that it was not recognized earlier would not defeat a defense of inherency. MEHL/Biophile Int’l Corp. v. Milgraum, 192 F.3d 1362, 1365 (Fed.Cir.1999); Atlas Powder Co. v. Ireco Inc., 190 F.3d 1342, 1348 (Fed.Cir.1999); Abbott Laboratories v. Geneva Pharmaceuticals, Inc., 51 U.S.P.Q.2d 1301, 1305-06 (N.D.Ill.1998), aff'd, 182 F.3d 1315 (Fed.Cir.1999). But Apotex has failed to prove this by clear and convincing evidence, as required to invalidate a patent. The serendipitous appearance of a polymorph years after another polymorph of the same crystal is patented does not prove to the requisite degree of certainty that the new poly-morph was inherent in the old. In this respect the uncertainties in the scientific community concerning the provenance and causality of polymorphs aid SmithKline because of the heavy burden of proving invalidity coupled with the narrowness with which the Federal Circuit defines inherency. Inherent anticipation “may not be established by probabilities or possibilities. The mere fact that a certain thing may result from a given set of circumstances is not sufficient.” MEHL/Biophile Int’l Corp. v. Milgraum, supra, 192 F.3d at 1365 (emphasis in original); see also In re Robertson, 169 F.3d 743, 745 (Fed.Cir.1999); Glaxo Inc. v. Novopharm Ltd., 52 F.3d 1043, 1047-48 (Fed.Cir.1995); Continental Can Co. USA, Inc. v. Monsanto Co., 948 F.2d 1264, 1268-69 (Fed.Cir.1991). I have been assuming so far that inherent anticipation requires that practicing the old patent invariably produce the new product (the product that is the subject of the challenged patent-here, the hemihyd-rate-and is contended to have been inherent in the practice of the old patent) from the moment the old patent was issued. This may be too strenuous a requirement, though I cannot find any case one way or the other on the point. Maybe it is enough if, because of widespread seeding, by the critical date for anticipation (October 1985, one year before SmithKline applied for patent 723, see 35 U.S.C. § 102(b)) no one who followed the steps outlined in patent 196 for making paroxe-tine would fail to produce at least some hemihydrate. And this is possible. But again the burden of proof on the issue of validity saves SmithKline. For it is equally possible, as far as anyone knows, that practicing patent 196 in non-seeded premises-and goodness knows there were some in the United States as of that date-would not have produced any hemihydrate. Cf. Glaxo, Inc. v. Novopharm Ltd., 830 F.Supp. 871, 875-77 (E.D.N.C.1993). Apotex has still another arrow in its invalidity quiver, however, and that is the argument that the description of claim 1 of patent 723, the only claim in issue, is too indefinite. A patent’s specification (the descriptive part of the patent) must “conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.” 35 U.S.C. § 112; see, e.g., Beachcombers v. WildeWood Creative Products, Inc., 31 F.3d 1154, 1158 (Fed.Cir.1994). Claim 1 states simply: “crystalline paroxetine hydrochloride hemihydrate.” SmithKline argues that this means what it says, so that even if Apotex’s generic paroxetine product contains only a single crystal of the hemihydrate, an undetectable quantity, claim 1 will be infringed. SmithKline correctly notes that the reference group for claim construction is limited to persons knowledgeable about the relevant technology-chemists in this case- and that the four words constituting claim 1 are each of them perfectly clear to chemists and that the four words taken together are also clear to them, at least as a matter of literal interpretation. Yet until completion of the portion of the trial devoted to the issue of validity Smith-Kline equivocated between the literal interpretation and a looser interpretation that would add the words “in a detectable amount” to the end of claim 1, or (a reading it preferred) “in a detectable amount whether or not detectable by the methods available when the invention was made or the patent applied for or granted.” Eventually, however, though reluctantly, it committed itself to the “single crystal” interpretation, according to which “crystalline paroxetine hydrochloride hemihydrate” includes all manifestations of the hemihyd-rate, no matter how or where produced, or in what quantity relative to the mixture of which it is a part; even if the production was inadvertent, unavoidable though undesired, and wholly without benefit to the producer or detriment to SmithKline in the sense of cutting into SmithKline’s market; and even if the amount is so tiny as to be beyond the limits of detection of any instrument present or foreseeable and the product in which it unexpectedly pops up does not compete with anything made or sold by SmithKline. It need not even be a drug. According to SmithKline, if a worker in a chocolate factory popped a Paxil into his mouth and as a result the factory became seeded with hemihydrate crystals, and the seeds found their way into chocolate that was then sold, the sale would infringe patent 723. (Actually this is wrong; the “first sale” doctrine would allow the worker to do anything he wanted with Paxil tablets that he had bought. The example is saved by assuming that the source of the seeds is a SmithKline executive who picked them up while visiting a seeded SmithKline facility and later visited the chocolate factory.) If the BCI factory is seeded, as it appears to be, all the other drugs that BCI manufactures there, some two dozen or so, may be seeded too and all are thus infringing (if SmithKline’s interpretation of claim 1 is correct and does not invalidate the claim) and the company has a slew of additional patent-infringement suits to bring. This is heady stuff; someone not steeped in patent law might think it loony; and it is not surprising that SmithKline equivocated before embracing the single-crystal theory (embracing it perhaps out of fear that otherwise its proof of infringement would fail-as indeed it would, as we’ll see). It continued until the end of the trial to duck and feint by arguing that whether patent 723 claims single crystals is “academic,” either because infringement cannot be proved if the infringing component of a mixture is below the limits of detection or because injunctive relief, as in my chocolate example, would be inequitable. The second point is a potentially fatal concession, as I shall point out later. The first is wrong. Being able to detect the presence of hemihydrate seeds by instruments and being able to infer their presence from circumstantial evidence are not the same. To equate them is a variant of Apotex’s erroneous belief that if seeds cannot be “seen” there is no scientific basis for supposing them, or the phenomenon of seeding, to exist. SmithKline insists that my chocolate hypothetical is “not this case.” It is not. But one tests a definition by considering all the things it covers, its “extension” as philosophers say. If someone challenges a definition of “swan” as any bird with a long neck by pointing out that flamingos have long necks, it is no answer that the person offering the definition is not interested in flamingos. It is still an overinclusive definition. SmithKline may not interpret its claim narrowly to avoid invalidity and broadly to prove infringement. Unique Concepts, Inc. v. Brown, 939 F.2d 1558, 1562 (Fed.Cir.1991). Reference to hypotheticals does not violate the principle that “claims may not be construed by reference to the accused device.” NeoMagic Corp. v. Trident Microsystems, Inc., 287 F.3d 1062, 1074 (Fed.Cir.2002). As the Federal Circuit had explained in an earlier case, “A claim is construed in the light of the claim language, the other claims, the prior art, the prosecution history, and the specification, not in light of the accused device .... [Cjlaims are not construed ‘to cover’ or ‘not to cover’ the accused device. That procedure would make infringement a matter of judicial whim. It is only after the claims have been construed without reference to the accused device that the claims, as so construed, are applied to the accused device to determine infringement.” SRI Int’l v. Matsushita Electric Corp. of America, 775 F.2d 1107, 1118 (Fed.Cir.1985) (en banc). In other words, the judge is not to say, “I can make everybody more or less happy by first examining the accused infringer’s invention and then narrowing the patentee’s claim so that it just excludes the defendant’s invention, leaving the patent claim otherwise intact.” But disallowing that method of claim construction does not exclude consideration of hypothetical cases as an aid to determining the scope of the claim. My objective is not to protect chocolate makers from being dragged into court by SmithKline (a most improbable prospect, . and one which SmithKline expressly disclaimed in its post-trial brief), but to decide whether the claim has limits that render it definite. The “single crystal” interpretation of claim 1 may be extravagant, but it is not completely ridiculous, whatever a layperson might think. The claim is of a product rather than of a method of production; and to make the patentability of a product depend on the percentage of the patented product in a mixture would lead to absurd results. The paroxetine in Paxil is mixed with excipients that constitute 90 percent of the pill by weight, yet no one supposes that the relative weight of the paroxetine in the pill has the slightest significance for patentability. Unlike copyright law, moreover, patent law does not recognize a defense of independent creation. The fact that if Apotex does create hemihydrate as a byproduct of its attempt to manufacture public-domain anhydrate it will not be deliberately “copying” SmithKline is irrelevant. But there are mixtures and there are mixtures. In a previous order in this case, which SmithKline applauds for holding that hemihydrate does not forfeit patent protection merely by being mixed with something else, Judge Kocoras stated: “It does not necessarily follow, however, as SmithKline would have us believe, that claim 1 covers any modicum of hemihyd-rate found in any mixture with other substances or polymorphic forms.” SmithKline Beecham Corp. v. Apotex Corp., No. 98 C 3952, 2001 U.S. Dist. LEXIS 19766, at *43 (N.D.Ill. Nov. 30, 2001). He went on to say that he thought the claim could be construed in a way that would preserve its validity when read in light of the patent specification, citing Exxon Research & Engineering Co. v. United States, 265 F.3d 1371, 1375 (Fed.Cir.2001). He pointed out that some of the properties specified, such as higher melting point, greater stability, and nonhygroscopieity, “distinguish the new material from prior art anhydrate and thus establish its patentability.” SmithKline Beecham Corp. v. Apotex Corp., supra, at *45. And he said that he thought there was enough information in the specification to be able to construct a definite meaning for claim 1 and thus save the patent’s validity. It falls to me to find that meaning. Id. at *47. Were claim 1 to be interpreted as broadly as SmithKline now contends, it would fail for indefiniteness. (I emphasize that this is not an issue that Judge Kocor-as even reached, let alone decided in SmithKline’s favor.) The purpose of the statutory requirement of definiteness is to place the members of the relevant technical community on notice of the scope of the patent so that they don’t infringe it inadvertently. General Electric Co. v. Wabash Appliance Corp., 304 U.S. 364, 369, 58 S.Ct. 899, 82 L.Ed. 1402 (1938); McClain v. Ortmayer, 141 U.S. 419, 424, 12 S.Ct. 76, 35 L.Ed. 800 (1891); All Dental Prodx, LLC v. Advantage Dental Products, Inc., 309 F.3d 774, 779-80 (Fed.Cir.2002); Solomon v. Kimberly-Clark Corp., 216 F.3d 1372, 1378-79 (Fed.Cir.2000). For remember that inadvertency is not a defense to infringement; there is no defense of independent creation. “The primary purpose of this requirement of definiteness in claims is to provide clear warning to others as to what constitutes infringement of the patent,” 3 Donald S. Chisum, Chisum on Patents § 8.03, p. 8-18 (2002), because “a zone of uncertainty which enterprise and experimentation may enter only at the risk of infringement claims would discourage invention only a little less than unequivocal foreclosure of the field.” United Carbon Co. v. Binney & Smith Co., 317 U.S. 228, 236, 63 S.Ct. 165, 87 L.Ed. 232 (1942), quoted in Exxon Research & Engineering Co. v. United States, supra, 265 F.3d at 1375. Stated differently, “if patent infringement can be compared to trespassing, the claims serve as the boundary markers that define what is, or is not, an encroachment on the inventor’s exclusive territory.” Alan L. Durham, Patent Law Essentials: A Concise Guide 66 (1999). This purpose cannot always be fulfilled .by a literal interpretation. The ordinary meaning of a word can deprive a claim of clarity. Northern Telecom Ltd. v. Samsung Electronics Co., 215 F.3d 1281, 1291 (Fed.Cir.2000). And context can ambiguate as well as disambiguate. Claim 1 must be read against the background of the specification. Solomon v. Kimberly-Clark Cory., supra, 216 F.3d at 1378; In re Moore, 58 C.C.P.A. 1042, 439 F.2d 1232, 1235 and n. 2 (Cust. & Pat. App.1971). SmithKline acknowledges in its post-trial brief that patent claims are to be interpreted in light of “the patent’s specification” as well as the claims language. Patent 723’s specification points out the superior handling characteristics of the hemihydrate in the manufacture of a paroxetine salt for therapeutic purposes. This would lead a reader, however technically adept he might be, to suppose that the patented invention was of a crystalline form constituting the active ingredient of a paroxetine-based drug. He would not think the claim extended to the involuntary creation of the crystalline form in quantities so minute as to be of no therapeutic, manufacturing, or other commercial significance and indeed to be undetectable by any means known at the time the patent was applied for and was issued. There is no hint of such a possibility in the patent; the landmine that SmithKline now wants to explode under Apotex is concealed. (Not concealed from Apotex, though, whose CEO expected from the start that SmithKline would sue it.) An adequate description of claim 1, if indeed the claim embraces even a single crystal created by adventitious seeding, would warn the world of such dangers as are conjured up by the chocolate example. The example may seem fantastic but the reality is alarming enough-as the ritonavir episode recounted by Dr. Bernstein (SmithKline’s own witness, remember) illustrates-for any firm contemplating manufacture of a c