false, You in fact deduced from the main conjecture , for a triangle, V~E+F = 1, But this we knew a ‘The method of analysis-synthesis* V-E+F=a forall polyhedra’-This ye rom tis we intend flat polygonal networks’: we noted that in polyhedra are simple’ that" V—E+F=1 for all inference we used also the lemma Qs that we simply-connected’. s we finally inferred Py that gle V-E+F = 1". And this trivial finding was accepted with 1 ANALYSIS-SYNTHESIS: A PATTERN OF EUCLIDEAN HEURISTIC AND ITS CRITICISM Prologue on analysis and synthesis : Pulleaie feats on mnder why. Because we arrived at something which is 1 Siould like to come back to your proof of the Descartes lubly true? But from false premises we can validly deduce true . igs CEE “ions: so we cannot conclude anything about the truth of the 5 You cla mises. Anyway we hnow that in our case all the premises are false. sree ou claimed that you proved the Descartes-Euler conjecture Uhr To be frank, Tam struck by your argument. have only simply conceseel ere ear2 are simple’ and ‘ll polyhedra Gamma: (But surely there is no real difficulty here.] ‘This chain of words, you in feet orien aces”. Though you did not puttin these Wlerences - I call itan ‘analysis’ ~ can be reversed and by this fact criticized those who thou a au fectire¥ and/sbewed istic ee ead cond prove tie ‘we can validly deduce P from the indubitably true premise P, and f certain subconjectures, The vem, your improved conjecture, was 1 This diagram may help you: nothing but a disguised inference: From the lemmas the oven follows." I admit that you adde: aera Teen fallow f that you added that this inference may be regarded Abiagih we stretch some of its concepts, but this i i You certainly claimed that your ‘proof’ was a ional theeriginal Sine four ‘proof’ was a deduction of the original q Pinan conjecture from certain lemmas not all of which may have been eee specified at bla: What are you driving at? Come to the point ~ if you have one Alpha: This invers renee eating back * This chapter (he i fer from our original inferences. For instance from P and Q, swe deren tie, Scion {wo paper writen Wee inferred P- But does the fact that we can infer P, from Pand Q, i Pis false, thesis written between 1936 and conference Jyrashya Pe (Gee Hindkka and Remes address were in the formnof Which occur between square brac {guarantee that we ca but P, true and Q, true, we cannot poss ly to infer the true P, from the false east aad typescript of Lakatos's Jyviskyl ‘even though we may be al ‘these points we have made various interpolations eae fhe soserilc ea See el and the true Qy SoU eaten se (Raine an fe facer re is coat Oe ad Os ae vrnad int petites which pk ut hove phedra 2 ey (This is essentially what Lakatos says in section Ta whieh the improved pro iy ofthe reverse inference from. this role when and only when QP.) However, Aipha's forrect condlsion seems to us invalid. For the same reasons ”‘THE METHOD OF ANALYSIS-SYNTHESIS theorem we must try to reverse this ‘proof” Psi: Indeed we may. Teacher: When your science teacher ‘proves’ you his scientific theo- ries by deducing undebated facts from them he follows this same pattern. I wonder why you do not protest to him too, Psi: We shall. Analysis-synthesis and heuristic separates the processes of finding the tru * But this does not exclude cither the process of discovery or of proof: Proof implies finding lemmas. But where do the lemmas come from? ve mind does not like proof which requires a jump to unknown lemmas ~even if th are the listed axioms of a il truth implies the dubitable ‘one? One would have to guess, to fall back on trial and error. But Primitive man shrinks from guessing. He abhors freedom, he feels ‘unsafe if he moves beyond the bounds of ritual. If he guesses, he does iouslyS Primitive men prefer decision-procedures. With the help of a decision-procedure one can decide mechanically whether aconjecture is true or false. Primitive men worship algorithms. Their concept of ce that of Leibniz, of Wittgenstein and of modern formalists, is essentially algorithmic. But the Greeks did not find a decision-procedure for their geometry, ough they certainly dreamt of one. They did, however, find a compromise solution; a heuristic procedure, which is not quite al rithmic, which does not always yield the desired result, but which is still a heuristic rule, a standard pattern of the logic of discovery, ‘This heuristic method was the method of analysis and synthesis. Let me state it as a rule: Rule of analysis and synthesis: Draw conclusions from your conjecture, one after the other, assuming that itis true. If you reach a false conclusion, then previous foot Father it thus the problem is notto infer (the adn textbooks and peri neither by Cauchy, 762, PP. 157-8 (ed, he two branches of Pappusian heuristic; ‘Gee p. 7). However, this separation in practice Ch Lakatos (1976, THE METHOD OF ANALYSIS-SYNTHESIS ur conjecture was false. If you reach an indubitably true conclusion, your Tovjcturemay hve ben ran thi case ever th proces ork backwards ‘and try to deduce your original conjecture via the inverse route from the indubitable truth to the dubitable conjecture. If you succeed, you have proved your conjecture, he first part was called the analysis, the second part the synthesis. BB curials rule thowa at once why the Greeks held rca od absurdum in such exceptionally high esteem: it saved them the labour of the synthesis, analysis alone having proved the case. a ‘The characterization of the method can be found in Euclid’s Ble ‘examples of analysis 1, The best preserved English re patho alysis then takes that which is sought as if it were admitted and Pasce through as socceive consequent the esta ses for aay vere (alread) done (yews), and we in fest and again hati the antecedent cau of he later and soon, unl Dyao racing our steps we come uponomething already krown or belonging 3 cs of fat principles, and such a method we ell nals cers (nine ulin sythens reversing te proces, we take a lead done that which was las avd atin the anya by arranging in thet natural or Tonsoquences what wercbefore antecedents ands we nrnve Beall at and aif ie were ex Inypoteis 0 som tha ih gh wl a be ea he problematical Kind we is propounded as if yar which we pn ough ies succesve consequences, taking them ar true, up to some Miniged: if then @) hat aimed pouible and basta, hat what athens lien, nat a im nd the proof will again correspond in Feverse order to {d we cote upon something admire impossible the problern imposible too were true and gedrated then yf hateomethingad tnd the proof wil corespond a 2B‘THE METHOD OF ANALYSIS-SYNTHESIS ‘This method has several peculiarities. One is that. culiarities. One is that a false conjectur can be disproved, but not improved by it. Anoth ied the only proofs that can be found by usin 1¢ one single axiom or a single proposition al not a serious restriction because axiom or proved proposi ice versa. But this is not always the case Some axioms eg may ena the conjecture under discus i conjecture may not entail the axioms. In such cases the method of analyiaynthoss docs net work: However we have wo le uatonest lure ofthe method in these cases? a tions of the method need: explanation. The Greeks must have come across los of theorems unprovable through analysis-synthesis (though the proportion of theorems with necessary and sufficient conditions is surprisingly high in the Element). Hankel has TThe theorem: "The vertices ofall triangles that their angles pes tly when the circle pases throu hand only fr those angler whi these cont a acirele,then EA: EB= ED: EF if A and B lie on the same and EF have the same or opposi ous then ED theorems of eg. the which we 8): “| have noticed two. paseages that may ‘would add THE METHOD OF ANALYSIS~SYNTHESIS \¢ embarrassed silence of the Greeks about stich failures was, I pose, at least partly due to the central doctrine of Aristotelian ism that genuine proofs (or explanations) have to be final and ich cohered so well ty were wholly ignorant of the analytic method, but, in my opinion, ‘poctuse they set so high a value on it that they wished to keep it to {themselves as an important secret.” (©) The Cartesian Circuit and its breakdown ‘The classical Euclidean programme is anti-empiricist; it is highly 1e senses. Indubitable propositions can be guaranteed only je intuition. Facts have to be proved from indubitable fir fhe method of analysis-synthesis may work quite Jn Euclidean geometry. Inmodern science however two new factors enter. One’ of indubitably true proposition: the reasoned fact. Reason un counter to sense-experience. They may ~ as Galileo put it~ the senses’. Examples of reasoned facts are: ‘the earth is round’, ‘all th the same acceleration in a vacuum’, and so on. The cs of proble possible under cerain volume Hust gropin oe babe was proving: thought be beneath one's digTHE METHOD OF ANALYsIS-syNTHESIS [Facts and reasoned facts are d facts do not direct! the entailment leductively unconnected, Reasoned ly entail occult hypotheses, though in this latter case n the opposite direction, from occult hypothesis to facts to occult hypotheses, from oces and vice versa has to be restored. We have kinds of statements to the same level of ct So I troducing a new sort of ve’ inference. [Thus the Pappusian cir- cuit mentioned above might be represented as in the diagram below } Aedsction This is an extended version of analysis-synthe arose from the Cartesian attempt toadapttheanci eee ipttoadaptthe ancientan, it has qui deduct The extension ralysis-synthesis nce. Tt is very different from the Pappusian schema: i empirical basic statements! and it has inductive as well as propositions thos propositions through whi Circuit, ee entice gp ‘THE METHOD OF ANALYSIS-SYNTHESIS ‘The thesis of this section is that a main feature of the story of modern scientific method is the critical elaboration of the ancient Pappusian, Gircuit into the Carte: followed ~ in spite of some partial successes and several intriguing rescue-operations ~ by its break- down. Let us first clarify some problems concerning the Cartesian Circuit. (ct). The Circuit is neither empiricist nor intllectuais. The source of knowledge isthe Circuit asa whole of the intellect. In fact there are very few, if any, pure-grained empiricists and pure-grained intellectualists. Descartes, Newton and Leibniz. certainly agreed that one can indubitably intuit truth andjor falsehood at both points: on the level of facts and on the level of first principles. Both may serve as basic statements, But everybody also agreed that ‘one cannot talk about true factual statements or true first principles taken in isolation; only fools trust sense-experience, and first principles from ont of the blue are just speculations - neither has a place in the perfect, infallible body of Scientific Knowledge. They are only respectable and suitable candidates for truth or falsehood if they are already embedded in the circulatory system of analysis-synthesis. "Basic statement’ is meaningless outside analysis-synthesis. Both Descartes and Newton were very explicit about “rash anticipation Some apparently should be interpreted in this light. For inst ‘And I have not named them hypotheses with any other object than that it may be known that while I consider mysel ry truths which I explained above, yet I particularly desired not to do 80, in order that certain persons may not for this reason take occasion to build Up some extravagant philosophic system on what they take to be my principles, and thus cause the blame to be put on me. ‘This passage corresponds very closely to Newton’s ‘Hypotheses non {fingo’. It means that hypotheses have to be embedded in a Cartesian Gircuit and thereby cease to be hypotheses. Descartes (1637) 29. The last worl reflec the effect of Gale's conviction. ”‘THE METHOD OF ANALYSIS-SYNTHESIS ‘The aim of the Cartesian Ci to carry truth to all the ' z r carry trut points in the Gircuit, thereby turning ses into facts, and justifying ian claim that the ‘conviction of pure science must be s The Gircuit does not allow unsubstantiated fantasies which would be inconsistent with the dignity of infallible science. That in this set-up effects and causes, and factsand theories, are on the same logical and therefore epistemological (though not heuristical) level (causa aequat effect) is also clear from the following passage from Clarke's Fifth Reply to Leibniz: the attraction, gravitation or tendency of bodies \ow sufficiently known by observations and experi- ments. If this or any other learned author can by the laws of mechanism explain these phacnomena, he will not only not be contradicted, but will moreover have the abundant thanks of the learned world. But, in the mean ‘ompare Gravitation, (which is a phaenomenon ot actial matter of Epicurus’s declination of atoms ( i facts, is an absolute rule for both Descartes and Newton. This has to be stressed repeatedly beca perian age which ov , ought ever to precede the method of compos the correct interpretation of the Fifth Rule tion’ Again Descartes's Regulae: he who would appres Glosely a he who enters ‘Theseus: Butmany people either do not reflect on the p 5 cprecepta it altogether, or presume not to need fe Consequently they often investigate the mos fic qcaon whole regard oder tha om min, ey ignorance ofthe nature ofthe heavens, and even without havi ‘made proper ooservations of the movements of the heavenly bodi “ to be able to indicate their effects This i also what Mechanics apart fromm Physics, ana rath set about de ir al al Line re Goal oe Rape ies ona ae In another passage, in Rule 1v, Descartes compares those who look nate, 73 7h Query 3 ‘THE METHOD OF ANALYSIS-SYNTHESIS for manifest Truth lying by the wayside withoutapplying the laborious ‘method of the Circuit with ‘a man burning with an unintelligent desire {o find treasure, who continuously roam[s] the streets, seeking to find Something that'a passer-by might have chanced to drop’. Descartes {Admits that this ‘method’ ~ pursued by ‘most Chemists, many Geo- metricians, and Philosophers’ ~ may lead to occasional windfalls. But fone has to pay dearly for them, since ‘unregulated inquiries and {fonfused reflections of this kind only confound the natural light and blind our mental powers’! This was Descartes’s real opinion of that brand of intellectualism which was later attributed to him! But while dismissing the use of First Principles in isolation from the Gircuit, both Descartes and Newton thought them to be an essential part of the Circuit which contributed decisively to the safety of this epistemological structure. Itis known that Newton was unhappy about the occult character of Gravitation and that he tried to deduce it - by the theory of ‘umbrella-effect’- from Cartesian First Principl Descartes sharply criticized Galileo for omitting First Pris ‘thus building without a foundation’ And he had a very reasonabl point, Facts alone are not reliable enough to guarantee the truth the Circuit. Mersenne and Rocco simply refused to accept Galileo's (Before the Atwood-machine facts about free fall were very on's work was hindered by false astronomical (ca) Indwel worth clarifying is the rel sian logic. Both are inferences based on ransmit truth [from premises to conclusions] and re- is made clear in Newton's famous letter to C For anything which isnot deduced from phenomena ought tobe called a Hypothesis, god hypotheses of this kind, whether metaphysical or physical, Wier of occu ualites or mechanical, have no place in experimental Philosophy fe this Philosophy, propositions are deduced fom phenomena, Mind afervard made general by induction? Or again, in the same letter, equating induction with deduction: experimental philosophy proceeds only upon phenomena and deduces general propositions from them only by induction’ * Descartes [ iabeeare 19‘THE METHOD OF ANALYSIS-SYNTHESIS ‘THE METHOD OF ANALYSIS~SYNTHESIS ly a discrepancy between formal Aristotelian logic, nineteen valid forms of inference, and induction which has ion. But Descartes brushes Aristotelian logic aside with contempt,! and replaces the poverty of sjllogistic logic by the infinite richness of intuitive deductions, the infallibility of which are guaran- teed by God. But if so, why should God not guarantee inductive inference just as He guarantees deductive inference? Another feature of the Cartesian Circuit which has been frequently misunderstood is the relative length and importance of deductive and inductive passages in the inferential chain which transmits truth from the facts to the occult hypotheses. ‘Newton was very keen indeed to deduce facts. In his priority quarrel Hooke only guessed the ‘ulmost completely forgotten. Only an occasional student of the history fof geometry remembers it.’ If a scholar stumbles upon it from time {o time and learns that the Greeks deduced mathematical theories from {ets (ie. axioms from conjectures), he is likely not to believe his eyes, phius F. M. Cornford not only thought, like Duhamel, that this method {is out of date, but he insisted that it is a paradoxical, ‘nonsensical’ the Greeks could not possibly have followed, and argued that Pappusian analysis is in fact identical with what Duhamel called the modern reductive analysis? According to Cornford, everybody Who interpreted Pappus in the deductive way, ‘misunderstood him Jamencably’. He says: “You cannot follow the same series of steps first -n the opposite way,and arrive at logical consequencesin both theory fully from the Hooke he repeatedly stressed that All proper proof Heluctant to notice the recalcitrant ca for theory from facts has been ridiculed by philosophers from Duhem onwards. The only physicist hho came to Newton's defence was Born, the first person in the history of science who reconstructed Newton's deduct ately Born missed one important does not and cani ion, but only to the inverse square la , but decisive inductive gap fi the inverse square law to the Universal Law of Gravitation, But this gap should not be overestimated. Newton almost deduced his theories from facts, and I should not be surprised if the same could be shown to be true of Planck's, Einstein's, or Schrédinger’s results too. Nowadays it has generally been assumed that while deduction leads from theories to facts, it has not the slightest part in the path from facts to theories. Both the Pappusian and the Cartesian Circuits have disappeared into oblivion. The last philosopher who took the Pappusian Circuit seriously enough to criticize it at all was J. M,C. Duhamel. He treated the ancient method with some contempt: as something out-of-date and vastly superseded. The modern method of | analysis is not deductive, he claims, but reductive, working from the follows, till we get Huh proof of the Euler-formula shows, itis still a major pattern of mathe- ‘matical heuristic. But let us come back to Descartes and Newton. Newton certainly argues in some places that he deduced his theories from facts. This can ‘wo equally plausible ways: first, that he thought that negligible; second, that the separation of the two ferences, deduction and induction, was ‘There is in fact nothing pi fof facts and theories in the Gartes (of its most obvious features, ‘obviously cannot mer ‘Newton meant that the prope meee Soar realized quite dearly ® Newion (1686), in Brewster This was po © Duhamel |B. 352) synthesis have epistemological relevance, did too, The source and guarantee of the tru 81‘but not how to attain certainty. “They did not seem in'whlch gronter ob i cently plain vo the mind itself why those things are so, ae i discovered them’.Phispassage containsa most crushing This isthe chief result which T have had in view in writing this treatise? luclidean synthetic stylet which, he says, stifles the mind: iles the mind, why then did Plato refuse te that ney 40 his School those who were’ unversed in Mathematics? dion confirmed Descarte suspicion that in antiquity ‘had knowledge of a species of Mathematics very different Which passes current in our time’. And it is on this crucial hhe in fact refers to Pappus ~and also to Diophantus, the BL Algebra: Indeed I seem to recognise certain trace of this Pappus and Diophantus'* BRP rea i croacne of tne Mathoss Unioralin Wy how it is that although people know the meaning and (ce of mathematics, they still neglect it while laboriously pur- ines which in fact depend on it? ‘are many other passages in Descartes in which he refers to is and, again, to Algebra as the starting point of his Discours for instance he repeats that he studied three the Method is the Pappusian one. The usually deal with problems as simple and unobstructed and stated with as perfect clarity, as those of Geometry. The reference to Algebra is worth explaining, Algebra is considered here not as a branch of Mathematics, but as @ Method, a twin to the method of analysis. And indeed, Algebra is in the Pappusian sens 1e connection all the other chem as though they were is rule. (Rule XVILF eater ccueears wel sera prey Coract only eats tac see aberars toch os appear ee eee Beebe eae ear sige Se eee ee canal eee eee ae clare A glimpse at Pappus’s classical text is enough to ascertain d tributes to its cultivation. ‘This made me fe Se Algebra is indeed "analye’- Dencares's veroce felons, 2 ae ye Pattern of problem-solving and not that of proof. Descartes explains that he was not interested in mathematics for its own sake: he was ted in the important secrets of the Universe and not in ‘the tr the reader who follows my dtife with suficient attention will easily see that nothing is less in tmy mind than ordinary mathematics, and that I am expounding quite another science, trations are rather the outer husk rematics mainly because ‘no ns of such self-evidence and fy Descartes. He found certainty in Bese * Di, p71 han by s “He dic not know that Buclid’s Blemen i Were mean tobe a cosmological theory sa: pe. 47-8) can © Bia. 84‘The essential identity of Rule 1v and this basic passage in the ‘Discours is striking: both of them contain Deseartes’s account of the ‘Three Sources of his Method. i Asiviote ie ald ihe bass of hls arguren of na yriori, showing the he ynd clear iby the senses and the traditions of the ancents. fat What you refer to is the method he uses in writing his doctrine, but te that with which he investigated it: Rather, {think it iby means of the senses, experiments and rive himself as much as possible of his conclusion. After- ignores the Pappusian tomake them demonstrable. Thats whatisdone forthe most is comes about becaise when the fs tue, one may by making use of analytic methods hit upon some BORE Ue akadY emetic or auvive at some axiomatic uti the conclusion i fase, one can go on forever without ever ding iced one docs not encounter some impossibility oF Archimedes and Apol To conclude: one = and Descartes’ Gireuit* (The Pappusian ranslated from Arabic into Latin by Com- mandinus in 1566 and by Halley in 1706 - was much discussed in the seventeenth century. It figures in Galileo's Dialogue on the Great World Systems: + -Theonly Cartesian scholar who seems io have appreciated the impact ofthe Pappusian Rober (1997), Untortinately even Robertiisinerpretsthis Descartess problem was t get tid of synthesis, to have an analysis which co ‘) in lgebra proofs were reversible, and (0) the algebraization of geometry and of the ‘ould therefore enable us to have perfect demonstrative analyses, To quote gr onl the iment of root bu prot nifest absurdity.* , let us quote Arnauld’s version: We may hence understand that this is the analysis of the geometers; for it on having been proposed to them, in relation | be a theorem, of its truth or falsehood: if ecessary consequence, they conclude Wrver and returning then through the way they had come, they demonstrate : Buti they fall, asa necessary to sone absurdity oF im- ‘proposed to them is false and though the myth of the reversibility of ‘what may be said generally touching analysis, which consists more {n judgment and sagacity of mind than in particular rules Jot necessary to investigate these more or less corrupted ns now. Our only point was to show that the Pappusian Gircuit topical in discussions of heuristic in the seventeenth cen- i was indeed part of advanced logic courses. Nor need we now (0 the question either of the continuity of the Pappusian tradition in medieval logic or of its place in the Paduan methodology.) But, granted that Descartes started with the Papj what did he preserve from it? Did he in fact develop it into the Cartesian Gi ry? js essay ~while leaving these questions open - we shall be content to claim that the Cartesian Circuit in fact és the rational Yeconstruction of che problem in question, and that history can only ly understood in the light of such reconstructions. ted Drake's translation, which only became hich Lakatos gave (el) ogulae, According to ‘what the ancients achieved ‘Algebra ‘designs toe the matter of figures’. This means that artes put TNigebra and Geometry on a par, And one may wonder whether his Rule xx does hot warn ts not t indulge in possibly irreversibe alge i ‘Rober’s original problem ~of which this sain his 2" Réponues dat both fei: the deep and atentive one here starts analysis with reasoned arts ~ unlike Pappu ~with basic analysis + Aenauld and Nicole (1702). 315 \ iments, wit, the first principles, (*We hav Robert and Brunsehvicg from the French (ds). 86 ‘ 87 ueTHE METHOD OF ANALYSIS-SYNTHESIS (4) The Cartesian Giveuit in mathematics ‘Those who - like Descartes ~ identified mathematics with Euclidean Geometry and Elementary Algebra thought that in mathematics facts are reasoned facts and occult hypotheses do not exis In the seventeenth and eighteenth centuries calculus, ‘unreasoned’ roblem. Cauchy and his followers solved the problem by the ‘translation-procedure* which corresponds to the inductive passage from facts to reasoned facts in the Cartesian Circuit. At the same time occult hypotheses also appeared. The explanation of some facts about the real line using ‘complex function theory is analogousto the transcendental hypotheses of physics. To deduce these hypothe of the problems which the arithmet Cartesian Circuit may bring to the surface some aspects of history and philosophy of mathematics which have hitherto been missed. (€5) The breakdown of the Cartesian Circuit () Induction does not transfer truth. One important stream of criticism was directed against the safety of the truth-transference along the inductive passages came under the attack, 10 the occult hypotheses. passage from reasoned facts to the: tities! denied value injected at the reasoned facts can ever reach the occult hypotheses. If we accept this criticism we can either (a) give up inf ‘confess the conjectural character of scientific hypotheses or (8) replace + GE. Lakatos (19704) p12 "The first rit Kneale (1949), pp. 97-8). Leibniz ence: He knew the condition rent from that used by geomet and incontestable principles, whe are tested by the consequences derived from them. The nature ofthe sujet ermits no other treatment ‘Newton wi by infallible Phenom THE METHOD OF ANALYSIS-SYNTHESIS this particular inductive passage by the infallible Method of Division.* Phere is a third, neutral way: (0 to introduce a theory of the prob- Ability of scientific hypotheses (which however leads inexorably to ly untenable theories of confirmation which try to reintroduce y through the back door)? vy of the Method of Division was crushingly criticized ¥y Catholic logicians from Pope Urban VIII to Duher® and by many the probabilistic theories of confirmation were crushingly zed by Popper. the course of these cr by Sbsinaiaewibadteyee Pee ich cbs called the ‘Newtonian deduce paage) bas been . of deducing theories the bathwater. in the seventeenth and especially in th as widespread in mathem: i) Improved deduction transfers truth perfec {oo were later sharply criticized, but never abandoned as conveyor! ‘They wereimproved in a piecemeal way by several trans es (arithmetization, set-theore Metho of Bivsion eg Dsante ig occult hypotheses wi the pessible conjectures from which the he watery element is observed recip ing some ng vessels, both of you would reply that He ould vr tis ays which are to some para dation ant regarded Tnalible, The Barth iter ‘destroyed one by ich sal Brat aequal defect o make = ghten the‘THE METHOD OF ANALYSIS-SYNTHESIS THE METHOD OF ANALYSIS-SYNTHESIS raised the level of rigour by reducing the number of quasi-logical constants, and at the end arrived at proofs which could be checked by Turing-machines. But one had to pay for each step which increased rigourin deduction by the introduction of a new and fallible translation. ‘The impact of this fact has not yet been sufficiently appreciated.* There are no fist principles and there are no perfect reasoned facts. ) The breakdown of inductive logic destroyed the [Cartesian] Circuit. in the truncated circuit flows only in one direction. The Braithwaitian zipfastener takes the place of the Cartesian Circui can the Braithwaitian admitted: yes. If n ‘e we may prefer to use the term ‘explan: liference seems important only to those lo first principles in mathematics and deny them chus use the proof versus explanation divide (.e. first prin- Jes versus no first principles divide), as a demarcation criterion ween mathematics and science: isan essential difference in the way in which we thnk by means ofa Whematical and by means ofa sciendfe deductive system, In the former Mart from the beginning and go on ply fe thc sere wart froe the Bi Gil logically, the epistemological order being fo its use again the snetaphor of But fastener transmit truth? If first principles are ing and go on to the end nly disprove. sian Circuit underwent asecond attack: not ing channels but jections. Cartesians wo levels: first principles and reasoned facts. the optimistic search for first principles whose truth will grow on us of human knowledge, in Economics (L. von Mises) , however, gen of conformity with experience) is assigned atthe bottom fist and then ing upwards...{I} tok a fong time for scientits to realte that the ico-deductive inductive method of science was epistemological dit- bf mathematies: and that, tem, they were not losophy of science ~ and indeed of we are ‘This logicist demarcation between science and mathematics is how. ‘ever unjustified. ‘The arithmetization of mathematics was the for the neo-Kantians’ claim that they could deduce from Peano's five synthetic a priori axioms. The logiciza- ton of mathematics was the basic argument for the logical positivists at they could deduce all mathematics from the analytically true axioms of the Principia. attempt was genuinely Gartesian.* Buthe failed. His axioms inity and choice were anything but anal fof some of the rest was also problematic, One can of course sti back on Kantianism, and claim that one's logical axioms are synt @ priori.® But this reinterpret weaknesses. One is following the forceful revival of the ancient Greek criticism of sense- ‘experience by Duhem and Popper ~ we can allow only tentative recog- nition to basic statements. The Popperian epistemological zipfastener, sm, is unsuitable even for le disproof. It cannot prove and itcan disprove only tentatively. As to the heuristic zipfastencr, this may start from a reasoned fact ~ like Kepler’s Law ~ may move upwards through a deductive passage and then make an inductive jump to the Theory of Gravitati then turn back along a purely deductive path, erase the former fact, write down the corrected Newtonian version of Kepler's Laws, and so on, There are no absol rerfectly reasoned facts. ym mathematics P30) our proof-procedure: Kepler's Laws as the pri ‘only in eliminating the Axiom of Reduc the Newtonian correction as the theorem. The © CE Lakatos (1g76), chapter 2 (ede) nus thought he had proved his law of inclined pressure, D, Bernoulli the para- ram of forces, Euler the principles of ‘and Lagrange the principle ual displacements. This was also Maupertus's motive in tying to redce Newtonian mechanies t inTHE METHOD OF ANALYSIS-SYNTHESIS that the trans! priori. The ot mn-procedures are again anything but synthetic @ hhat the Gédelian argument about incompleteness Of reazonably rch theories shows that the inflible air of arithmetic fact resently known part is 01 fraction ofthe infinite whole, E au ‘One could argue that the Gédelian proofs of undecidability which lead to these new axioms be perfectly simple and translucent. destroyed also these Hilbertian dreams about tesian dream of the trivialization of knowledge in science but also in logic and mathematics." consequences. Exactly the 3 logic, the logically fet pri ne logy, Russell a Cartesian logic. Al approach the origin * We did not discuss here the ‘THE METHOD OF ANALYSIS-SYNTHESIS 2 ANALYSIS-SYNTHESIS: HOW FAILED ATTEMPTS AT REFUTATIONS MAY BE HEURISTIC STARTING POINTS OF RESEARCH PROGRAMMES lessor Hintikka’s and Mr Remes's essay is an interesting contribu- (o the vast literature on the problem of Pappusian analysis~ thesis, and 1 am delighted to have the opportunity to supplement the rather different views which I advocated in my ‘Proofs and futations’ and which seem to have escaped their attention.’ (a) An analysis-synthesis in topology which does not prove what it sets out to prove ke Polya, whose work unfortunately is ignored by Hintikka, regard as a heuristic pattern which although it may have been started reeks has been characteristic of scientific and of mathematical ch up to the present day. ling you of two classical examples of analysis. polyhedra, Euler, in 1751, proved that forall polyhedra V-E+ F= 2, jere V is the number of vertices, E is the number of edges and F number of faces of the polyhedron, 's proof ran as follows. Let us assume that in fact it is the | V-E+F=2 for all polyhedra. Let us then take a particular of a polyhedron, for instance, the cube, and perform on it wing experiment, We first prepare a hollow rubber model of a cube, with the edges painted in red.* If we cut out one of the faces, we can stretch waining surface flat on the blackboard, without tearing it. The +s and edges will be deformed, the red-painted edges may become ved, but V, E and F will not alter, so that V-E+F-=2 for the iginal polyhedron if and only if V-E+F= 1 for this flat network remember that we have removed one face. (Figure 1 shows the flat, wr the case of a cube.) Step 2: Now we triangulate our map i We draw (possibly linear) polygons which ‘not already (possibly curvilinear) triangles. By drawing each dia- increase both E and F by one, so that the total V—E+F will sred (Figure 2). Step 9: From the triangulated network we British Journal for Tunder the tide: lish as Prefs end ther information on my ‘nd this volume, & ‘The description of Cauchy's proof | pp. 7-8. (Eds)‘THE METHOD OF ANALYSIS~SYNTHESIS THE METHOD OF ANALYSIS-SYNTHESIS now remove the triangles one by one. To remove a triangle we either blackboard. We need to assume that all flat networks at which remove an edge upon which one face and one edge disappear ive in this way can be triangulated without changing V-E+F. ire 3a), or we remove two edges and a vertex ~ upon which one yeed to assume that from all such networks all triangles can be face, two edges and one vertex disappear (Figure 38). Thu: id one by one until we have reached the last one again without V-E+F=1 before a triangle is removed, it remains so after the wing V-E+F. Our deductive chain is really more like triangle is removed. At the end of this procedure we get a single triangle. For this V-E+F= 1 holds true, I a, a (ee Ee) ——_____—> ET») \¢ derivation of a very weak conclusion (V-E+F=1 for our Je) was reached from a strong premise only with the help of some Strong assumptions. However, once we make use of these strong jumptions, we can work our way backwards, from the triangle to earl polyhedron and derive Euler's theorem from the fact that a Ingle has 3 vertices, 3 edges and 1 face. The analysis provides the synthesis. The analysiscontains the itive innovation, the synthesis is a routine task for a schoolboy Is case the creative innovation was the idea that polyhedra are ‘re ed, triangulated, rubber surfaces. The analysis, incidentally, was formed on one specific polyhedron, and therefore the universal lemmas were only suggested but not made explicit. However, the hidden lemmas are false. Notall polyhedra are homeo- Wworphic with the sphere and not all polyhedral faces are simply nected. Therefore only those polyhedra are Eulerian whi y assumptions. Both the analysis and the synthesis are the theorem which we set out to prove turns out to be a re conjecture’, But nevertheless we can extricate from the xr from the synthesis) a *proof-generated theorem’ by ig the conditions articulated in the lemmas. Thus we don nut to prove V-E+F=2 for mn of theanalysisand a more where P, is a special polyhedron (namely the cube) and Tp, is the it synthesis we do n starting point. triangle resulting from the transformation described in the “proof”. start with the proposit el “The predlsette serra rari Eulerian thelr eae aI re eet { process of imaginative-critical analysis-synthesis we arrive at the quasicEulerian (i.e. the property which holds for those objects for proposition *All Cauchy polyhedra are Eulerian’. which V-E+F= 1). What happens then to non-Cavchy polyhedra? This problem en- However, this trivial derivation strongly suggests the more general indered a veritable research programme, It led to a full classification formula e eo tye valent coseinnrfaces totheclaseicationof tuply av) ——> BT), connected polygon sets and to the calculation of V-E+F for a wide “range of topological objects, In the course of this investigation a series where P is a free variable ranging over all polyhedra. In this case, fof further hidden lemmas emerged and finally the research as a proof of Euler's conjecture since all that Cauchy did was to prove that if V-E+F= 2 for a cube, V-E+F=1 for a specific triangle. But the latter equation is surely trivially true. However, this curious proof hhas immense heuristic power. We may of course describe the inference which it involves by however, we need very strong auxiliary assumptions to derive from srogramme evolved a hard core (in the axioms of algebraic topology) E(P) the conclusion ET»), We need to assume that all rubber poly- Inet sophisticated, rich postive heuristic hedra after having a face removed can be stretched, without tear, flat But if we start from a proposition P and draw consequences from it 95,Ie Pe N THEE SYNTHESIS ‘THE METHOD OF ANALYS! it follows, then wi lest rather than prove. ‘Thus he ica inwise ~a conjecture; but if we fa jere is then a pattern by which one gets from naive Popperian ing to the method of proofs and refutations (not conjectures and | and then, one step furl hematical research . This pattern refutes the philosophical claim that the | source of research programmes is always some big meta- ‘A research programme may be of humbler origin: it izations. My case-study in a sense frequently the study of facts and practice of low-level generalizations which serve as the launching for programmes. Mathematics and science are importantly in- d by facts, factual generalizations and then by this imaginative luctive analysis."* turning it hrs into a proof and then into a matheveatied research programe ae «this point we mightraise the problem of how we arrived origi at V=E+ Fea, at the ‘naive conjecture” tant teas th mort mathematical conjectires appear if are usually proved before the axiomatic system is articulated in the proof canbe performed ina formalized wi hed matical conjectures by tentatively solving Obl m rroblems by tial and eror. Thus we raise the problem ast the relatons betceg the faces, edges and vertices of a polyhedron. We try out different conjectures one after the other. I described in detail, following Polya, this series of conjectures and refutations.* It took in this case nearly : ‘onjecture by Popperian conjectures an ave" period the first tage of mathematical i lar case from Euclid to Descartes. But at some stage the naive Part en occ titan age ae Iysis and synthesis starts: this is the second stage of discovery which | called ‘proof-procedure’, This proof-procedure generates fer the brand-new proof-generated theorem and then a rich research pro. gramme. The maive conjecture disappears, the. prookgenersted ® become ever more complex andthe cente ofthe tage d wented lemmas, fist as hidden (enthymemes) and later as increasingly well articulated au ; : these hidden lemmas which, finaly, becom: reat mathe (0) An analysis-synthesis in physics which does not ‘explain what it set out to explain should like to switch briefly to a second example. Before I do so, should like to draw your attention to the fact that at least up to the luge that I carried it, there was nothing specifically mathematical in iny first example. Everything that I said can be interpreted as a | Yeiearch programme in networks painted on closed rubber sheets. In | “this case our analysis leads to an explanation rather than to a proof,and | ihe emerging hidden premises are explanatory propositions. I repeat: this case we would have deduced from Euler's formula its own explanation. sis clearly the casc in Newton's celebrated analysis-synthesis which an explanation-procedure father than a proof-procedure. But before turning to Newton let me ‘empha: V—E+F=2 for all polyhedra’ is no less and no more Janets move in ellipses’. low-level hypotheses: Kepler's three laws of planetar He took ~as one does in analysis ~ one instance of 4h planetary system: one in which tl is held in a fixed ps by an invisible hand and in which there is only one planet orbiting It: He setoutto perform an ‘analysis’ of Kepler's laws for this particular ‘ase, First he deduced, in his chosen particular instance, the purely Kinematic implication that a plane planetary motion has eration directed toward the Sun, from Kepler's nai the radius vector covers equal areas in eq) ‘ease, this end-result about the acceleration is not evidently true, but it certainly has a degree of plausibility in the light of Platonic meta- physics. Then Newton proceeded to the synthesis. Assuming that the acceleration of the plane motion is directed towards the Sun, he or of a set of lemmas known to be true, We start wit conjecture and we have to aera conceptual framework in wl we find that in a heurist Jemmas vent the lemmas, nd even perhaps the the lemmas can be framed. Moreover ly fruitful analysis most of the hidden found on examination to be fale, and even known to at the time of their concepti very different from Hintikka’s (or Pappus's) conception of (theoretical) analysis. In my conception the problem is n rove a proposition from lemmas * See Lakatos (19760), chapters and 2 (eds) 96