You are on page 1of 6
FUZZY RESOLUTION PRINCIPLE Zuliang Shen Liya Ding Masao Mukaidono Faculty of Engineering, Meiji Unversity 1-1 Higashi, Tarek, Kawasaki, Japan 214 ‘ABSTRACT ‘This paper will extend te resolution principle based pon reducto ad absurdum from binary logic ino fuzzy logic. The extension Is dependent on introducing the concepts of fuzzy Contradictory, contradictory degree, fuzzy resolvent and ontidenceoffevolvent In ths paper, we shall rst interpret the ‘Concept of fuzzy resolvent because iis 4 mos concept fa a hse furry eatin. Then we sl onl he {izzy resolution principle for the forzy propositional logic at first and extend it to the fuzzy fust-order logic. The Completeness ofthe fuzzy resolution principle wil be proved at last Based on the fazzy resolution priclple, a powerful fuzzy prolog system hasbeen established 1. INTRODUCTION Maybe 1965 was one ofthe important years in he history cof harman thought From Aristotle to now. Ta this year maior trealthoogh was made by J. A. Robinson, who inoduced the soll reoluon prt. Another major reaktrough war made by L.A. Zadeh, who introduced the concept of icy int et iory and li Since 1965, many sient have {eying to find sealing met in fery loi as ffectve {5 the Yesoltion principe i binary log, and some of them Such as R.C.T. Lee (1972 JD) and M. Makaiono ( 1982 1043 JO" have found cern inveresing properties of esolton in fuzzy loge, Unfortunatly, the esostion principe i based ree of redo estan whow dep bate the ‘Somplementry aw in binary ogi, but the complemen faws do nu hol in furzy fot. Maybe for this reason, thee have Been no deep-ging discostons p10. now about how to Use hea of redat ad sand for fanny inference "Ins paper ol inode sore import concep suchas the fuzzy Contradictory and contradictory degre, the fuzzy ‘erolven and onfidence of solvent, et. By these concept, at {tcompleterefationl (or zy reftauonal procedure whch it dopendent on so-called fuzzy resolution principle can be Esstlched. Though the premises and conclusions, swells {he inferential ress se incomplete (or fuzzy inthe T0229 ‘efatational procedure, we can prove th the fuzy etuation [procedure based on he assy reason pine complet. It [te fuzzy refraonal procedure, the tth values of premise {ed oncto ae ltaken erm ese (01 then he ey ‘solution principle is equivalent to the binary reseltion Flociple, Therefore the binary resolution principle can be Tooted pon as special as of uy resaiaion principle. (0196-628%88/0000/0210801.00 © 1988 IEEE 210 2. FUZZY RESOLVENT Fuzzy logic can be defined as an algebraic sytem <( 0, TLA.V., > where the closed interval ( 0,1] isa set of truth values (inthis paper, the logical ruth value is from the concept of maltipe-valued logic, but it can be easily extended {othe linguisteal uth value for the fuzzy semantics set fom Zadeh ) and where logical operations AND( A ), OR( V ) and [NOT(™ ) are defined as follows: AAB=min( A,B), aay AV B=max(A.B), (22) A (ornot(A))=1-A, ABE[Q1). (23) A variable C1 = 1p on 2) oF ts neaton 5 sdf be tert, nd ad se si 0 be Complement ofeach ther os pairof complementary vate ‘X'loue a disjneton of eras, or formula consiing of ORY of some er ‘When itis convenient, we shall regard ase of tral as synonymous with a clause. For instance, { x1, x2. X3 mV A clause consisting of n literals sealed an ‘clause, A one-literal clause i called a uni claue, When a clase ‘omains no literal, we eall it the empryclauze Since the empty ‘lause has no lite! that canbe stifed by an interpretation, the mpky cause is always false. We customarily denote it by [ and T(L]) © 0 forall interretstios. A set S of clauses is regarded af a conjunction of all clauses in S, where every ‘arable in' is considered governed bya univers quantifier Definition 22: as follows! Fuzzy formulas are defined recursively 1. A variable nis. fzzy formula 2.IEF isa fuzy formula then F isa fuzzy formula S.1FF and G are fuzzy formula, then F A G and F V G are fuzzy formulas. 4. The above are the only fuzzy formulas. Giving an interpretation I he wath valve ofa clause Ci determined uniquely by aubsttuing a vale ofthe close interval 11, 1] determine by te interpretation | fr each variable ofthe clause. Tats, an interpretation is mapping from each variable to the set ofthe uth values [0,1]. We wil wrte the truth ‘aloe ofa clause Cunder a given inerpretaion Tas TC). For 4 set 9 of clauses, we will write similarly the truth value of $ under a given imerpretation a5 Ti(S), where if $= ( Ce p Cale then Tx) means TS) TCA. ACR) f= min( TC) TH(Ca))- ay In fuzzy ogi, the complementary laws (1( 2) V 1x) 1, TCX) A T(x) = 0 forall interpretations ) do no hod, Thus, a clause in which xj and Xj are involved simltaneoosly is significant in fuzzy logic. Hereafter, we will call such as clause a complementary cause: Definition 2.3: Leta pair of complementary variables x; ‘and be under a given imterpretaton I. Then x; is said 10 bea contradiction under a given interpretation 1. IF Tix; A %i) then itis sad wo be complete conradictary.1°T}¢x) A ¥) '5,then tis said tobe non-contradictory. I Ty(xj A Xi) € (0,05), then iis sud to be incomplete contradictory Definition 2.4 ‘contradiction xj A % be Let the contradictory degree of a d(x) . . = max (Tx) THR) mine TH), THE ) Qs) under a given interprecation I. Whee itis evident hat ed(x;) © [0,1 iseld forall interpretations bv contin i comptes (a sonoma ) forall nterpreatons i and onl ie {oneal dogeeis atte Cor) foal nerremtons Fordseaton, when we dicus a contradiction in inary he concep of consadiory degre ca be neglected since ft stsape pie condita. ant when we dicts a ontadicton in fszy loge, fs Contain degre sua 0 Tw be meanings for refurational inference because itis son-conraditry Definition 2.5: Let the truth value of a contradiction x, A 5 be equal othe “ruth value of its contradictory degree (e4( i) which s defined as fellows: Ted) Hed(x1)) 05.405 rrin( TH TH) ) ATO AR) 26) forall interpeerations. Where, itis evident shat Ted 34) € [0, (05 hold forall interpretations. Definition 2.6: Consider two clauses Ci, C2, Crem VL, GV Le @ay where Land L2 donot contain te tera or Xj asa factor and have no pair of complementary variables. Then, the clause Ly V Lis suid to be a resolvent of Cy and Co whose keyword is x: td the contzadictory degree of the KeyWord is €@{ 3) A resolvent of Cy and C2 is writen as RCC}, Cp), and a fuzcy ‘resolvent of Cand Cp is writen a8 RC Cy, Ca Jed, where ed = au ‘sth contradictory degre ofthe keyword a is sald to be the confidence ofresoient of R(C},C2)- Definition 2.7: Let the uth value ofa fuzzy resolvent RCC), Cade Be TER(CH. C2 ea) =TR(CH.C2))V Teed) (2.8) fo al interpretations. In binary logic, because the confidence of resolvent eis always equal {and T ed )is always equal 00, (ed) can be neglected, tats, T(RC C3, C2 ea) ® TCR(C3, C2): When A, B, C, D € [0, 1, the following well-known inference rls nods ponens: fA and A ~* B, then Bi ‘modus tollens: if B and A ~ B, then A; disuneive syllogism: A V Band, ten B; hypolheicl syllogism if A~ Band B ~ C,then A~* Cs consuucive dilemma f(A +B) A(C-*D) tod AV C.then BD; and desuvive dilemma: if( A+ B)A (C~ D) and BV B, thea VC SHAE Bs he hany ese RCEh, Coen frm two rdecestr clauses under he defition of he pico a8 AwBHAVB, Qs) ‘The above special cases hold, for instance, because in modus ponens A and A ~ B is equal to AV (AV B ) under the definition (formula 2.9) ofthe implication -> and also ‘because its fuzzy resolvent is B (with its confidence of resolvent ed = ed( A) = max( A, A )~ ming A, A) ), and by sme reason, of them can be binary logic Is Used ina problem-solving system, one ‘sores a statement A instead of ifthe woth value ofA i 1. (IF ‘he truth value of a statement A is 0, one simply stores A.) In fuzzy logic, we should store a statement A, instead of A if the ‘ruth valu of Ais greater than or equal o that ofA. Tats, we store Aif (A) 2 1-T(A). In this ease, TA) 2 0. se game hone ied satay in zy logicas follows. Definition 2.8: An interpretation Lis suid 1 satisfy a fonnala $i TS ) 2 0.5. An interpretation Tis said to falsify S WTS) £05. According to the above definition, if TS ) = 0.5, then T both satiaies and flsifies 8 That i, in the St of sh value (0, 1", iis considered that 0 and’ Ihave different definite informations and that ambiguity reaches its maximum at 0.5. $0 the auth value 0.5 is sad to be & meaningless potnt in our fuzzy Inference, Definition 2.9: A formula is sud wo be unsatsfable and only if tis falsified by all ts imterpretaons. ‘When applying logic w problem-solving systems, because ivoften is based upon the so-called reductio ad absurdum, we have wo prove the unsistabilty of format "Te following postulton isan interesting and important condtonforogic lence. ee Posulaion 2.1( Mukaidono, 1982 J: We postulate thara logical consequence Q from &premice Pi significant oaly ifthe fllowing condition i satisfied TP) TQ), thats, T(P)=T(PAQ) forall interpretations ‘The above postulation is easonable, because any logical consequence in binary loge statis the condition (210.10 fact, a resolvent RCC}, Co ) of causes Cy and Can be considered to be logical consequence fom the premises Cy and ‘Grin binary loge inte sense of (210); that the folowing ‘ondhion is said in binary ogi: TC) A Ca) < T(R(Ch. C2) ), thats, TELA C)=NEACARCC)). which means that the rath vale ofthe clases docs not vary for ‘Blierrettons even if te esovent of he ates aed ta facy logs, we can prove that the above elation always holds by inrodaeing the concep of a fizzy resolvent. (210) ‘Theorem2.1: Let Cy and Ca be two clauses and R( Cy, Ca eatay be fuzzy resolvent of Cj and C2 whose keyword is 1 Th the following relations hold: TCL A Ca) TREC. Ca ead) qa) TECLA Ca) = TCL A CA RCC. Ce ean) (2.12) Proof: We an asume without los of generality that Cy=xiV Lyand Co = iV La, where Ly and L2 do not include 1 and and where Ly V La does not inlade a Complement. Then, TECLAC2) *THARV AA LV HAL V LAL). TCR(CH. Ca ead) * TRC. 2) V The)? TL Va) V THAR) forall merges, Her, the following relations we always ‘ali forall interpretations COTO ALa) eT Vita), (THAT) Ta Vlad. G)TUA Ata) WV ia). “Then, by taking the spremum (V of both sides of the above thee relaons, we deive Tou ALAV HALA V Li A12) 6 TUL VL). ‘Ading (3A 5 1 bh sides by OR operation, we nd tat ‘Re reation (2.11) hols for al ineetations. And is evident that he tlaon (212) hos too when the relation (23) ola QED. a2 Corollary 2.1( Mukaidono, 1982): Let Cy and Co be two clauses and R((Cy, Cp be a resolvent of Cj and C2 whose Keyword is. Then the flowing relations hold? (ETOWAH) ETRECH.C2)). 213) then T(C1 A C2) ¢ TCR(Ch. C2) (2)# TCG A H)2 TREC, C2)), (aay then T(C, A C2) 2 TCR(Ch,C2)), Proof: We ean assume without loss of generality that iV Ly and Ca= ¥V La, where Ly and La do not include xj and ¥j and where Ly V Lz does not include a complementary itera The the prof of The 21, ne TA LV HAL V LA La) s TL Va). ‘Therefore, if Twi A ¥) § TCL V Lz) = TCC, C2) ), ‘then T( Cy A C2) TCR( Che Ce) and TEx A ¥) 2 TCL V La )= TRC CIs Ce ) I then TH Cy A Ca)2 TREC @)). QED. By Corolary 2.1, we know that if we use onl the concept of resolvent then it depends on the ith vale ofthe Keyword tthe resolvent whether or not we obtain a significant logis] onsequence In Postion 2 by taking a resolvent ia fey foes Mredecpy. we wil had tha ifthe curva of & resolvent bigger than or equa 005, then the resolvent has Say sigiicnce in Postlation 2.1, butt evident hat such Sth value of resolvent soften meaningless for problem. feivng systems because these systema are Based pn the So. Called feduetio ad absundu. Fortnately, onthe otber hand by {he ntodicing new the fuzzy resolvent in tis pape, Theorem 21 Shows vx hat te ory eeaivent as always algnifeance Postaltton 21. Tet Sb set of clause. Then, the set which const of and al forgy resolvent derived from any pi of clases of § ad which denoted as RICS eat (as) fs called the frst class fuzey resolution se of S, where ed is Said 0 be the firs lass confidence of resolvent and ed = ed, ‘Where ed! isthe minimal confidence of resolvents of Rha i ed = min( ed ed"... e41™), The m-th class resolution set, ofS, denoced as| RECS Jam (26) fs defined as OS a (ary ACS Jen = RUC ROS Jeg )ad (218) where the O-th class confidence of resolvens ed = 1, andthe rth las confidence of resolvents ed" = min( ed, eda) Lemma 2.1: Let Gleg’ and G2ga" be two fuzzy resolvents and R(G1, G2 de bes fuzzy resslvent of Gleg and G2cq" whose keyword is xy and ed = min( ed, e@",ed(x))). “Then, the folowing relations hal: Te Glea A G2eqr) < T(R(G1, 2a). 1 Glee A Grea") (as) (220) = T(Gleg A G2ger A R(G1,G2 eg). roof: We can assume that Gleg = (xi V Lt Je and (Geq= (is V La eat without loss of generality, where Land do not include x and % and where Ly V Le doesnot include ‘complementary literal. Then, ‘TGlee) = TCL) V Kee), ‘Te G2qar)=T(G2) V Hed" TGtea A Crear) (G1 A.G2)V (G1) A Tea") V 1G2) A Teed )V Te A Ted") and by Definition 24 and 25, we can show that Tied) (imine ded x;))) max( Teed), Ted"), Ted(x:))) Ted) V Ted") V Ted) and T(R(G1,C2)ea) T(R(GI,G2)) V Med) = T(R(GI, G2)) V Ted) V Teed) V Ted) {or all interpretations. Here, the following relations are always valid forall interpretations (CL) NGI) A Tea") eT"), (2) G2) AT(ed) ST), (3) Ted) A Ted") Ted) V Teed" ) and by Theorem 2.1, (4) TCGI A G2) T(R(GI, G2)) V Ted x1). ‘Then, by taking the supremum (V ) of both sides ofthe above our relations, we derive thatthe relation (219) holds fo all inerpretations. Ad its evident that the relation (220 ) holds {op when the relation (2.19 y holds QED. ‘Theorem 22: Let § be a set of clauses and Geq be clause of RY(S Jean Then forall clauses Gig of RMS Jas, TS) ST Gea), Proof: “IF Ggg is an clement ofS, then iis evident (in ‘he case ed = 1and'Y( ed) =O). Let us suppose that Ge. an clement of RICS Jea fesolvent from two cla Jn the set ofthe Oth class confidence of resolvent, sed = ‘min( ed? ed( x; )) and iti evident that ed’ and c40" are all larger than or equl to d®, Therefore, by the definition ofr lass fuzzy resolution set and Lemma 2.1, Clea A C2céo") -ST( Cea’) and T(R S Jat) = 1(S ). Next, lt Cog be an clement of Ri(S Jegi — R™-H(S Jeg. Then, in the similar manner, We can show that there are two clauses Cleat)’ 26d (CReainy in REIS ait such that Cori fzzy resolvent of Cleat) ui iy" Whose keyword is x3, Where, ed is ‘one eleent in the sot othe th clas confidence of reslvents and ed are elements inthe set of the (i-I)th las coatdence of resolvent, 50 ei» ma ed ell x)). ‘Then we have TC Clegg-ty A C2ea-y") $M Cea tat CS dedi) S TC Cea) and TC RICS Jeg) = ‘TUR(S jog), Therefore, tng Cog = Ge, we can ave athe heat 118) ST Gea) GED. aa Corolary 2.2( Lee, 1972)¢__ Let S bea st of clauses Let Ch. Cyt be clauses in S. Let max( TCG) }. TCC2 ). sn TUG )) = Band mine TEC) }, TECa). os TOM) =a 305. Let C* denote any clause inthe set RCS. Then forall n 2OasTO) sb. Proof: Because T(S ) means min( T(Ct }, C2. 4 Gy) by Theorem 2.2, there is TCS) <1 Cregg) = T( C8) V T( ed), and because T( $)=a>0.5, Teedn") s 0.5, therefore, a ¢ T( C*), On the other hand, if C is a resolvent of Cy!! and Cai-t, then it always holds that max( Cit, Cor) 2 CL Therefore, (C4) sb. QED. From Corollary 22, we know that fase of clauses S is not unstsfable, then we can do significant infereace in fuzzy oie: Obviously, this result shows us that we can not use fuzzy Inference to prove the unstiafiaility of set of elausesS. But ‘ow we ean se Theorem 2.2 to prove such unsasfiaility. So "Theorem 2.2 is one ofthe main results derived inthis pape. 3. FUZZY RESOLUTION PRINCIPLE FOR THE. ‘PROPOSITIONAL LOGIC The exenlea of he fury setaution principle to peck wher sero dames conte empty Sata) 1 SSSnais [ten $s unsisblef S doer conn] the nx tg to check is wheter jean be derived fom Sf ICS Ja telus oly an empty ase} wh is ofiece eet) SSRIS te oy cai a Sis saled fey refutation oa fey bro of 8 skh Sonfdence of rewolven €6" Evident, te confidence of ‘esovent | 2ea.20.1ed =, the ference rel sal Be ‘icant ht Teg) = Osc ork words now-coucltry. ted = te inference read be Meant ai TC (Jca)= 0, orn oer wordy, is ‘ciple consadon. BS genera, fra rel Seiler meaningless armenia atn,03 TU eg) =O, Stinciber wot hte ical onan “Sen te ccoce posed sid tbe tn icomplae ronal (or ‘sy rational) procedure by he ary rslaton ple Apeadin on scaled reaicn a ebsurdan if weave ome fry anny daz orm F representing contusion, ten we eta fom a he frzy {oral rom te conjtcon ofthe ions wi he teased ferzy formula (F), We can ten conver hs frm ino 4 number of clase in non fom, tat i he set of clase Uiing se iden ote" ey reilation principle, we ean tepeelly dv ww ei of ies agi eomesoences of (slg ont ith fy orm debe rom be tos Shen whl evenly be abet dae (pm n ih an pe ince {Toy te ingot foal pope af clan SILC fy fan compte po ce seston 4) From he gota couhdete of ented min cl, > Ean) whch Siw tate cial fray formal ve oe Shhh ust how mac eee fos hao Example 3.1: Prove that a formula A -* C can be deduced from the axioms A -+ B and B -* C, where TCA ) = 08,1( 8) =07 and TC) =09. Proof (it A+ Be AVBandB>Ce BYC, (2) ete conjunction of the axioms with A A.C (the negation of A~* C) be (AVB)A(BVC)A AAG then there ita st of causes S=((AVB)(BVC)A, 0). (BRA Sat = Dag wee smin( cd( A), ed(B )e6(C)) smin( max(T(A),TA))— min(T(A), TCA), smax( T(B),T(B))— min(T(B), 718), max( T(C),T(0))= rmin(T(C),7(2))) 04 hich sth confidence of resolvent (4) A~ Cho wit is confidence of resolvent 0.4 QED. Obviously, the resolution principe in binary lope is a special ease (ed = 1 ofthe fay resoition principe. In order to change a predicate into a proposion, cach individual variable of the predicate must be bound this lone in two ways The fit way fo bind a nal vra tyrasignng's value wi foe second tnehod of binding Invi vail by quandicaion ofthe vale. The most omanon fonts of guandeaon se wvergl and exe "By the approaches of subsiion and nication of binary Jogi we can easily extend the fuzzy resolution penile no fiatond loi Example 32: Suppose that we have the following sxioms: fa man hopes to mary a woman andthe woman hopes ‘omarry the man, then a mariage willbe eanged between the ‘man and the woman. Now, lt us find the magnitude of the Possibility that a marriage will be arranged between Mr. A (Cwhose hope is nt so stong, maybe only 6O% ) and Miss B (whose hope is stronger than Mr. A's maybe 80% ) Capa yen pron x hope o mary (ies mariage = 7) the possibility th mariage Xs nt the possibility that a reptsent the possibilty dy. Co) ERE AB) 06 re i sti ace hep eee kta pera may Bk) 08 resent te pol tak Miss B hope to ME AWEOR. (4) Wehavenw one ie und two fet as flows Temary( x.y) A tary yx)» managed x9) 2 mary A.B and TN many( A.B)) = 08 imam B,A) and T many By A)} = 08 4. eamage AB find miage A,B) = 1 (3) As in Beimple 3.1, e have a set Of clauses as Blows $= ( mary( A,B), mary B, A), ot( mariage A,B) ), not marzy(x, y}) V not mary y,)) V marriage( x,y) ) (6 )By the approuches of substitution and wnification, as wel as zzy resolution principle the empty cause [] With its confidence of Fesolvent ed = 0.2 can be deducted. This means that we ean derive that martiage( A,B )'is true with the confidence of resolvent 03. a6 4, COMPLETENESS OF THE FUZZY RESOLUTION ‘PRINCIPLE, Before we ty to prove the completeness of the fozzy resolation principle, some important results have t0 be Tnmodaced. Lemma 41( Lee, 1972 ): A set § of clauses is ‘unsatsfiable in f022yTogic if ahd only i its unsaisiable in binary ogi. Lemma 4.2( Robinson, 1965): A set of clauses is nsatsfiablein binary logic if and if there is a deduction of the empty clause [fom S. “Theorem 4.1( Completeness of the Fuzzy Resolution Pincple any causes is unstable if nd only i there is & eduction ofthe empty clase [] witht confers eset el = Of Suppore Sis snsatisiable. Then it is onsastabl in binary bye by Lemma 4 By Leta 4.2, then empty clause (Just be deduced rom S. And it i evident that by Definition 2.3 and 24, ed O08 T(J) < 05 always hols because ed = 015 a mzanngless point fr refuatonsl Inference. ee ‘Convery, supose ther is a deduction of | Jeg and ed = 0, Assume § i saiable. Then TCS.) 2 03 > Lea ). However, hii imposible beraus by Theorem 2.2, we have Showa tat 03 > RC{ lea) 2 TCS ) Hence, 8’ must be Unease QED. By Theorem 4.1, we know tat ifthe meaningless point 115) 2 Sif omit en he fotyvetaonl froetae. which is base ay esau pnp comple. ‘Piste anoer min esl deve nhs ape 5. CONCLUSION “The main result ofthis paper is thatthe fuzzy resoltion Fc mae comple y ining ce 2 egy een y he fey Tao nce cn rome ny proponna ‘Etccively asin binary loge ‘some special terms ae changed, we find tht the s0- called Musidono's Fuzzy Resolution Prnciplel®), which is based onthe 3-4 Makaidono's Postlation) is equivalent the fuzay resolution principle, Soin his paper, we Bave proved the completeness of the Mokaidono's Fuazy Resolution Prnsple to. ased on the fozzy resolution principle, we have established a powerful fz) prolog syste [REFERENCES (1) M.Mukaidono, 2. Shen & L. Ding, Forz Prolog, Preprints ofthe 2nd TFSA Congress, 7(198 [21 Z. Shea, L. Ding & M. Makaidono, A Theoretical Framework of Fuzzy Prolog Machine, Fuzzy Computer, dit by M. M. Gupta & 7. Yamakawa, North Holland, ‘Amsterdam, 1988. (3) LA Robinson, A Machine Oriented Loge Based on the ‘Resolution Principle, ACM, 12,1, pp. 23-41(1963). [4] L.A, Zadeh, Fuzzy Sets, Information and Control, 8, 3, pp. 336-35301965), (5) EC. T Lee, Fozzy Logic and the Resolution Principle J ACM, 19, 1, pp. 109-11901972)- "Nkaidono (6) M 4H. Masuzawa, Some of Resolvents in Fusty Logie, ans. 1ECE Japan (Glapanese), Vo. 166:D, No.7, 101983) (71 Me Mitiono, Fz nfscice of Reston Sse Fuzzy Set and Possibility Theory, Bait by R. R. Yager, ‘pp. 224-231, Pergamon Press, New York(1982). (81 Ni, Nilsson, Problem Solving Methods in Arificial Inieligene, McGraw-Hill, New ¥ork(1971). [9] CoLeChang, RC. T. Lee, Symbolic Logic and “Mechanical Fheorem Proving, Section 33, 5c pp. 76 80, Academic Press, New York and London(1973)

You might also like