Professional Documents
Culture Documents
E. A. Maxwell - An Analytical Calculus, Volume 3. 3-Cambridge (1966)
E. A. Maxwell - An Analytical Calculus, Volume 3. 3-Cambridge (1966)
E. A. Maxwell - An Analytical Calculus, Volume 3. 3-Cambridge (1966)
0, we have the relation ao ale a2 7m wae 4 again agreeing with the calculated result. CARTESIAN AND POLAR COORDINATES 21 To illustrate partial differentiation with respect to r, let us con- sider the coefficient = (fg. 111). This requires a variation from P to P’ in which 4 remains constant, so that P’ lies along the radius OP. Itis then an easy matter in elementary trigonometry to derive, for increments dz, ér, the relation so that, as dr>0, Fig. 111. Finally, consider, for partial differentiation with respect to 0, the coefficient z. This requires a variation from P to P’ in which r remains constant, so that P’ lieson the circle of centre O and radius OP (fig. 112), Draw PM, P'M' perpendicular to the axis OX, and PU perpendicular to P’M’. Then the increments 49, dz are given by the relations 80 = ZPOP’, -éx = M'M, the minus sign arising since éz is negative. The length of PP’ is 60, and the radius OP is at right angles to the are PP’. For the interpretation, we suppose that P’ is very near to P, so that PP’ is almost a straight line. Then 2UPP' = 3n-LUPO = yr-8.22 PARTIAL DIFFERENTIATION uP , But ZF = cos UPP’, = so that 52 = sind approximately, Honce in the limit, as 80->0, & = —rsing, as calculated. Fig, 112. EXAMPLES V Illustrate graphically the relations ar 02 1 ga sind. 2 ae ey _y by _ 3. ad. 4. 5h = 10080, 9. Connected ‘independent’ variables. Tt often happens that a function arty) is given in terms of two variables x, y, but that those variables are in fact related, each being a function of a single variable t (which might, in a special case, be « or y). Then zis a function of t, say z=F(t). [For example, the point (x,y) might be restricted to lie on a curve, such as the parabola = af, y = 2at.] CONNECTED ‘INDEPENDENT’ VARIABLES 23 To find an expression for the differential coefficient aor "(O), assuming the existence and continuity of the coefficients = Le and : de dy the existence of T, St. Suppose that t receives an increment u, as a result of which x, y become z+h, y+k. Then _flethy+h) fey) w —Pethyt+h-fey+th}+(feyth fey} u Maer Ohiy +h) +i i(e.y tok) 7 , as for the similar argument on p. 11, Hence Fetw= FEAF Bees Osby +B) +E file,y+ Oh). u de dy Now proceed to the limit, letting u tend to zero. Since 77, ot exist, tin 2, im, wot dt’ mcg also h, k both tend to zero with u. oe by the usual theorems on sums and produets of limits (Vol. I, p. 25) and the continuity of ee = Zz , the relation becomes FF AF pete iyen, aF Ode , a dy Ged ey at ‘This relation is also written in the form de _dede , te dy Gi” bedt * dy at’ where it is understood that z is expressed in terms of ¢ on the left and of z, y on the right. or, on rewriting,24 PARTIAL DIFFERENTIATION Intvsrrazton 4, Suppose that z is the square of the distance of a point P(«,y) on the ellipse oy ate from the focus (ae, 0). Then 2= (w—ae)t+y%. ‘The variables x, y are not independent, but can be expressed in terms of a parameter t in the form x=acost, y= bsint. Then 88 _ 9(@—as) = 20(008?—e), ee z= 2y = 2%bsint, fag Bd tt ay Gan asint, Ff = boost. Hence, by the formula, & = —20%(cost—e)sint+26%sint cost = 2atesint—2(a2— 64) sint cost = 2atesint— 20%" sint cost, For example, the value of z is a maximum or minimum when lz ‘ 7 0; that is, when int = 0, or cost = Ie, ‘The latter solution is impossible (since ¢ <1), and so the only turning values occur when sin ¢ = 0; that is, when P is at an end of the major axis. Intusrration 5. Suppose that z=f(x,y), where @, y are linear functions of the variable ¢; say, a=at+pt, y=b+qt. dz_tdz, ‘Then Bra astay x a te =P tts THE CHAIN RULE 25 EXAMPLES VI Evaluate F"(t) by the rule given in this section, and verify your answer by calculating F(?) explicitly and then differentiating it, for each of the following examples: 1. f@y)=atys =P yatt 2. flwy)=2ev%; e=cost, y=sint. 3. f(x,y) =2+y; w=cosi, y=sint. 4. f@z,y)=2e%; zat, y=. 10. The chain rule. The work of the preceding paragraph may be generalized. Suppose that the two* variables z, y are each functions of the three* variables £, 9, ¢; say =(, 7, ¢), y(E,7, ¢). Then a given function 1(w,y) of the variables x, y can be expressed in terms of &, 7, € as a function p(é,7, ¢). We obtain an expression for the partial differential coefficient oe ae in terms of the partial differential coefficients of w, x, y. When £, 7, £ receive increments 8£, 7, 8, the variables x, y become x+ dz, y + dy respectively, and w(x, y) becomes w(x + dx, y + dy). ‘The increment in w (or p) is (compare p. 11) w(w-+ dx, y + dy) — wl, y) = drwy +O, dx, y+ dy) + dy wi(x,y+0,0y). (1) Moreover, «, y are themselves functions of &, 7, ¢, so that the corresponding increments are drs o(6 +58, +89, 6+ 80) —2(E, 0, £) = dEa(E + 6,58, +87, 6+06) +9924 (8,9 + $299, 6+ 06) + Ca(E, 7, 0+ 9386), (2) with a similar expression in y. * The numbers, two, three, are only chosen for illustration and may be any Positive integers. 3 en26 PARTIAL DIFFERENTIATION On substituting from equation (2), and the corresponding equa- tion with y, into equation (1), we obtain as the coeflicient of 4 in the increment of w the expression wy +, dx, y + dy) a(E +b, 08,9 +89, f+ 86) +wy(w,y + Ody) Yes + WE, 7 +39, 6 +8), where the number y; replaces the ¢, of equation (2). Now restrict our attention to those cases in which the partial differential coefficients are continuous. Then (p. 13) "| . aw wi{a+ 0,8x,y+ dy) = wiley) te =a + ep ew wy (2, y+ O,8y) = wi(z,y) +e, ine E+ 9,88, 9+ 89,6400) = HE Otay = Set Ay % YEE + Wr 9E, 9 +39, 6+86) = Yi(E,9, 0) Ag = Mad, where Ay, A, tend to zero as 8£, 39, 8£ tend to zero; so that dx, dy also tend to zero, and hence ¢,, ¢ also. Thus the increment in w (or, to be more exact, in p, which is the same as w but with , 7, as the independent variables) is dp, where o- (SEs a) “(atm)” plus terms which vanish with 8, 89, 8. Hence we have the relations, forming the chain rule, tp _ dwae , dway a6 bx of * Gy 2? ap _ twee , cway THE CHAIN RULE 27 More generelly: Tf isa function of the m variables 2,25.» y which are themselves {functions of the n variables £, Ey, ..-yEqs and if w when expressed in Lorms of Ey, £,) «sy i8 a function p, then BF (ee 51,9, 450) t= SE Gant Innustration 6, Let w=ay%?, where smcosdsing, yaaingsing, 2— cong. dw _ dw | way , Owe 8D = dx 00" dy 20° Be 00 = — y's sind sing + 2xy2? cos 0 sing + 3my"2*.0 = —sin®d sin®¢ 005° +2 cos? sind sin® g cost = (2—3sin®0)sin@ sin® ¢ cos* 4; dw _ dwae , way , dw Be 26 bap * by Op * Ge O6 = y"2*cos 0 cos} + 2xyz*sin 0 cos $— 3xy%2"sin g = sin? 0050 sin®¢ cos's + 2sin® cos 0 sin®¢ costs —3sin?d cos sin*g cos? = 3sin®d cos sin®g cos* ¢(cos*g—sin®d). EXAMPLES VII 1. Given that w= 2y2, where a=cosOsing, y=sindsing, z= cosd, aw ew evaluate wo oe 2. Given that w= wot, where waatytz, v= rye, aw dw bw evaluate ie ty’ ae 3, Given that w = cos», where usaty, v=a-%, aw ew evaluate ie? Oy"28 PARTIAL DIFFERENTIATION 4. Given that w =a logyz, where z=u+o, y=uty, z=u-2, ow bw evaluate Ra’ Bo 11. Differentials. We now transfer our attention from the approximate theory of small increments to a more exact theory based on the concept of differentials. Suppose that 2ayte9) is a function of the two independent variables z, y, and that these variables receive increments dz, dy not necessarily small. Since we know what is meant by the partial differential coefficients = = (sce § 4), we may form the function ae, a et ay which, for given increments, has a definite value. It is called the DIFFERENTIAL of z for the increments dz, dy, and is denoted by the symbol dz, so that, as a matter of definition, a, oe d= meta In particular, we require meanings for the differentials of 2, y themselves. If we regard x as a function of the two independent variables 2, y, then a, y ea ay 0, so that dz, =1.d2+0.dy me Similarly dy = dy. ‘Thus the differentials of the two functions «, y for the increments dx, dy are just dx, dy respectively. Hence for given increments éx, by the differentials dz, dy, dz are connected by the (exact) relation aus dz = Fedo dy. Returning for a moment to the geometrical illustration of §6 (pp. 13-16), we can give an intuitive interpretation of the differ- entials, though a precise statement would be hard at present. The DIFFERENTIALS 29 surface (fig. 107) z = f(x,y) meets the vertical lines BA, P’Q’ at the points B, P’; suppose that the ‘tangent plane’ at P meets them in B,P’. The sections of the tangent plane in the easterly and northerly directions really are straight lines, and their gradients are exactly & & ae ay calculated at P. Hence the rise from P to P' is precisely a, | 2 att yh or, by definition, the differential dz, Thus the differential de is the increment in 2 when the surface 2 = f(a,y) is replaced by the tangent plane to it at P. For small increments, the two values are approxi- mately equal. 12. The differential of a function of two DEPENDENT variables, in the form dz = = ds+ % dy. With the notation of §9 (p. 22), multiply the relation proved where aay) de_tde dy Ged ya? by the differential dt. Then dz, tzde, dzdy au "matt wai dt, Now z (on the left), z, y are functions of the single variable ¢, and we know (Vol. I, p. 43) that their differentials dz, dx, dy are defined* by the relations da de dem Gat, dea Z dt dx dy dz . : cs y waa differential coefficients with respect to t. Hence the relation assumes the simple form zee Ya, wi e 2 de = = d+ ay giving the connexion between the differentials dx, dy, dz. * Te should be clearly understood that we aro appealing directly to the defini- tions. We are not morely cancelling df, though the notation dz/dt for a differential coefficient may create an optical illusion to that effect.30 PARTIAL DIFFERENTIATION The significance of this result will be appreciated on referring to $11 (p. 28), where precisely this formula appears under the hypo- thesis that the variables «, y are independent. The work of the two paragraphs thus unites to obtain the important theorem that, if zis @ function of the two variables x, y, then the relation ee ca dz= he d+ ay is true both when the variables x, y on the right are independent, and when they are dependent. 13. Digression on implicit functions. A detailed study of the properties of implicit functions is difficult, and cannot be attempted here. The present paragraph seeks to illustrate, mainly by example, the ideas which the reader should have at the back of his mind in the work which follows. Consider the relation ay — Tt can be used in three ways: (i) to express # as a function of y, z in the form way; (ii) to express y as a function of z, x in the form yartet, (iii) to express z as a function of z, y in the form zaartyt, ‘These three functional relations are implicit in the given equation, and we have, in fact, been able to make them explicit by direct solution, There are, of course, many implicit relations for which direct solution is (in practice, if not in theory) quite impossible; for example em —sin (x+y +2) = In any event, there are often advantages in keeping the equation in its implicit form, without direct solution, and so the corre- sponding rules for partial differentiation must be formulated. Returning to the example xyz = 1, we obtain the differential relations corresponding to the solutions (i), (ii), (iii) in the forms de = — Byte Sdy — by-%z*dz, dy = —$2tetde—kete tae, de = — forty tde— forty ty, respectively. IMPLICIT FUNCTIONS 31 Multiply the first equation on the left by a? and on the right by its equivalent y*25; then wlde = —3y-ldy —52-1dz, Multiply the second equation on the left by 3y-! and on the right by its equivalent 32824; then 8yMdy = —524dz—a-dz, ‘Multiply the third equation on the left by 52~* and on the right by its equivalent wty?; then Betde = —w-tde—B8y-Ady. It is immediately obvious that the three differential relations just obtained are identical. Indeed, they could have been obtained at once by writing down the differential du of the function us 2y'S-1 inthe form du = y5dx + Sxy*25dy + Say*Adz, and noting that du = 0 since w has the constant value zero. This example illustrates a technique which we shall adopt without further comment. Suppose that, say, three variables 2, y, 2 are connected implicitly by a functional relation ua,y,2) = 0. ‘Then we obtain the same relation between the differentials dx, dy, dz whether we find it directly by equating du to zero or by first expressing # as a function of y, z (or y of z, x; or z of 2, y). 14, Differentials of functionally related variables. Suppose that three variables x, y, z are connected by some relation, so that cither may be expressed as a function of the two others, Regarding as a function of y, z, we have dz = Saye eae, Tegarding y as a function of z, x, we have aa ae <—— dz+ en das Tegarding z as a function of z, y, we have & a ey Ye (As we have just explained, these three forms of the relation are in fact identical, the ratios dx: dy : dz being the same in each.) de = Fae32 PARTIAL DIFFERENTIATION An alternative notation for the partial differential coefficients is sometimes found useful as giving greater precision. We write, for example, ee a, to denote the result of differentiating z partially with respect to y when the other independent variable is z. With such notation, the three relations given above are em 5) +) a), a d=) drs) / a), * ey) [Thus if the implicit relation is e+yP+e=1, we Bb Bs Bat Bb ‘The three ways of looking at the functional relation, and con- sequent three ways of looking at the differential relation, all result in a linear equation among the differentials dz, dy, dz; and it is often convenient to express this linear relation with symmetric notation udx+vdy+wdz = 0, (where the symbols w, v, w denote certain functions of the variables 2, y, 2) leaving unspecified at this stage which of the variables are to be regarded as independent and which as dependent. With this notation, the partial differential coefficients are Gz vox) w chee ey) w oy uw Beans o 2 re a ute: &),"~w" &),7 we" DIFFERENTIALS OF RELATED VARIABLES 38 This leads to another observation. Consider, for example, the coefficient 20) ii) It is calculated by differentiation on the assumption that 2 (the other independent variable) is constant, and this assumption may be stated in the equivalent form that dz is zero. But then the differential relation ude +vdy +wdz = 0 becomes simply udz+vdy =0, and it follows, since a) _ that es) =de+dy, ey). t Oa) In other words, the partial differential coeficient =) , where zis kept constant, is equal to the ratio dx ~ dy of the dient iat de, dy when dz is equated to zero. ‘These ideas may be extended to any number of variables, and the resulting theorems are important in practice: (i) Ifzis eae of the acai By gy oop %q, then &z a= Zine sitet “eee te : a whether the variables 2;, 2g, ...,%, are independent or not. (i) If there is a functional relation between m variables YoYo --+r Yoo then their differentials are correspondingly connected by a linear relation of the form Uy dy, + Ugdyy +... +U,dy, or Sudy,=0, t where ty, tt, ...; Uy are functions of the n variables. Further, if we regard, say, y, as a function of the remaining variables ys, Ys, /,,, then the partial differential coefficient (calculated with ys, 44) -+»Yq all constant) is equal to the ratio dy, + dy,34 PARTIAL DIFFERENTIATION calculated with dy, = dy, = ... = dy, = 0; 80 that Wes yy Intustration 7. Three variables x, y, z are such that each may be regarded as a function of the other two. To prove that @.e)-» ‘here, for exarnmla, &) means thats expressed as a function of y,2, and then z kept constant in the differentiation. A relation between 2, y, z implies a linear relation between their differentials. Thus functions a, 6, ¢ of the variables exist such that ade+bdy+cdz =0. Hence @), =de+dy when dz=0 =-b/a, and (2) = ay+ae when de=0 =-a), so ths ().€), In the same way, @. onary (@)--« (= wm @,6),6).-—» Innustraioy 8. Four variables w,1,p,v have the property that each one of them can be expressed as a function of any two of the others. Terni) - 8) OG) ILLUSTRATIONS ON RELATED VARIABLES 35 where, for example, () means that 1 ia expressed as a function of t,p, and then p kept constant in the differentiation, ‘A relation between three variables implies a linear relation between their differentials. Since u, t, p are related, functions a, b, ¢ of the variables exist such that adu+bdt+cdp =0. ‘Thus @) =du+dt whon dp=0 ? = — bla. ‘There is also a relation between w, t, », and functions 4, B, C of the variables therefore exist such that Adu+Bdt+Cdv = 0. ‘Thus ) =du+dt when dv=0 beer =du+dv when dt=0 and This accounts for the first three partial differential coefficients in the relation which we have to prove. There remains @) , for which we need the relation between dv, dt, dp. This is found by climinating du between the two differential relations, giving bdt+edp _ di _BUu+Cdv @ = or aCdv+(aB-bA)dt—cAdp = 0. Hence &) =do+ds when, dp=0. ° b4-aB_bA B Salim 1a02 0. 80 that36 PARTIAL DIFFERENTIATION EXAMPLES VIII 1, Prove that, if u, y are each functions of the two variables x, t, and if wis expressed (by elimination of t) as a function of x, y, ei i) =m) /%) Illustrate this result by considering the cases @) w=et+e, y=at; (ii) w=eFcost, y= e*sint. 2. Each of the variables 1, », wis a function of the two variables x,y. By elimination of «, y, the variable w is expressed as a function of v, w, Prove that Illustrate this result by considering the cases @ waetty, veat—y%, w= dy, Gi) w=e+y3, verty, w=at—ayty® 15. Small increments. The example which follows is typical of many: Intusrration 9. The sides b, ¢ and the angle A of a triangle ABO are given by mfing omsins Ameo" A number of children (not knowing these figures) make measurements, with a maximum error of 2% in length and 3% in angle. To find the greatest consequent error in their estimates of the area of the triangle. The area z is given by the formula z= Yesin A. When dealing with products, it is often convenient to begin by taking logarithms, so we write uzlogz = log}+logb+loge+logsin A. SMALL INCREMENTS 37 ‘Suppose that the errors are 4b, dc, 3A, leading to an error dz in 2 or du in logz. Then we have the approximate formulae au = Mas, ou ou ou ou =H ta 3 tage so that Zar = Fa0+2 2 5e4 S04 aa, Now the coefficients 1/b, 1/e, cos A/sin A are all positive (an im- portant point) and so the greatest error arises when db, dc, 0A are all positive and at their greatest values. But we then have 3b = 20/100, de = 2/100, 6A = 34/100, so that 14, 1/2) ,1( 20) , cod (34 z aE 100) * sin. (100, 3A cot A | “a 100 Now the angle A is measured (as in all work for calculus) in radians, 80 that A = j7; also ra 3. Hence Ly 4 ,mv8 _ 84743 z 0 200 200° (847 y3)z so that 2 ‘The greatest percentage error in the estimate of z is thus H8+7 V3) % = 68% approximately. The corresponding error in actual value is 4x 4x 5x sin 30° x th or 0-34 sq.in. EXAMPLES IX Repeat the work of this illustration (with appropriate modifica- tions) for the following sets of measurements: 1, b=2, c=3, A= 45°, 2 b=4, c=5, A=150% 3. b=2, c=3, A= 135%38 PARTIAL DIFFERENTIATION 16. Partial differential coefficients of higher order. If z=f(w,y), then es are also functions of the two variables x, y, and possess their own differential coefficients. We uso notation such as a (2) _o% a) a (2\ _@% ze (ea) "a EG) =a § (8) "aa z= (ax+by)*, S = salaz+ by) & = sar +by) and so on, For example, if then Oe 2 Fan Oar +by), a san irene aa Fes Nan + by). Before going further, we give a proof (under the simplest con- ditions only) that the order in which partial differentiations are performed is irrelevant. In other words, we prove that, in ‘ normal? ie oe te Budy ~ dyox" [The proof which follows may be omitted at a first reading,] Let zaf(e,y) be a given function of the two variables x, y, and consider the expression Hafethytk)—fethy)-fey+h+fey). COEFFICIENTS OF HIGHER ORDER 39 First we express Z in the form E={f(e+h,yth)—ferh,y}-Pey+h—-Sey}, and then write ue) =fe,y +k) fey), where y is regarded as constant. Thus Esu(e+h)—u(2), and we may apply the mean-value theorem (Vol. I, p. 61) for the function u(c) of the single variable z. Hence E=hu'(2+0,h), where 0, lies between 0, 1, the differentiation of u being performed on the assumption that y is constant. Thus, by definition of x, Bali fifet Oh,y+k)—fie+Oh,y)}. vy) =fi(e+O,h,y), where «+0, is now regarded as constant. Then B=hwo(y+k)—vy)}, and we may apply the mean-value theorem for the function v(y) of the single variable y. Hence E=hkv'(y+9,h), where 0, lies between 0, 1, the differentiation of v being performed on the assumption that « is constant. Thus, by definition of v, E=shkfi.(0+0,hy+Ok), the differentiations of f(x,y) being performed first with respect to and then with respect to y. In the same way, we could have written Z in the form Balferhy+h)—-feyt+b}—ferhy) fey} and so obtained the formula E=khfi,(a+Osh,yt+Ok), where 0, 0, lie between 0, 1 and the differentiations of f(x,y) are performed first with respect to y and then with respect to «, Hence 5 (0+0,h,y + Oak) = fiy(a+0gh,y + Oh). Now let h,k->0 independently. On the assumption that the two second-order partial differential coefficients are both con- tinuous, we reach, as the limiting form of the equation, the formula Pf af yan” Bucy’ Now write40 PARTIAL DIFFERENTIATION Finally, we remark that, once this result is established, we can easily prove more general results such as Os Oia). he atdy — Gadyon ~ ayex®” and so on. EXAMPLES X Oe Oe Oe Pe Oe J Evaluate 5 gray’ Og Batay’ Bepyp OT cach of the following functions: 1. aty’, 2. aty', 3. e*siny. 4, logy. 5. wsiny. 6. we, 7. aly. 8. ytante, 9. atyter, Intustration 10. To prove that, if 2= xply/x)+ Wyle), ae a 40% then at arog tH aya = where f(y/x), ¥r(y/2) are functions of the variable y/z. Let ¢’, Y be written to denote ¢'(y/z), ¥(y/2), the differential coefficients of ¢, yr with respect to the variable y/z. (In other words, if t is written for yjz, then 9’ is TF with y/e written for ¢ after differentiation.) Then Zum =92ue =-$'ule, Zou = 4 2012) wy ay =n, with similar results for y. Thus, by the definition of 2, &_ yey vy eae oo? ee _ ao! eae wy 2 ts Ane et ILLUSTRATION: CHANGE OF VARIABLE 41 Ca Hence a5 Yay =x. Differentiate with respect to 2, y respectively. Then ee oe oat * ont Y aeay ee =9-P'y|x; de Oe ap arog TY ay dy = Py =¢’. Multiply these equations by 2, y and add. Then az cd ic A Bait aay tM at ae Voy =gte or, using the previous result, ae ae Oe a + Qay ty = 0. Inzusrration 11. (Change of variable.) To prove that, if ae bat = By? then the function 2 must be of the form z=fet+y)+oe—y), where f, g are arbitrary functions of their arguments, Write E=nty, y=a-y. Then Be BIR QP ge ag i gle and - 2 By? VAP QR RP AP RP AF giv42 PARTIAL DIFFERENTIATION It follows that the ‘operation’ a zl on a function of a, y is equivalent to the ‘operation’ a3 eet 6° en on the same function expressed in terms of £, 7; and that a &y Fae a is equivalent to aa Oz _ 8 (dz tus %-2(2) =a") (renusne ) @ 8\ (a, ee oF ~ (era) Gra) (retain 3) = 2(%, #)+2 +5) 36 ag Gy) * on ee” oy =e 2 4 + EE tpt Rey cine en Be _ de, Be tte ay BE OER ye Hence the given relation ag a (by definition) is equivalent to eee Thus so that = is independent of £ and therefore a function of y only; say B= May. Integrating, a fun ant, ILLUSTRATION: USE OF OPERATORS 43 where A is constant with respect to 7, but may involve Writing fun dy = 9(n) and A=f(), we have the relation 2 = f(£)+9(7), or, in terms of a,y, z=f(e+y)+g(e—y). Innusrration 12. (Change of variable by the use of operators.) Giren at u=e-y, v= 2Qry, 1 (2, 9 comes aig top) with u, v as independent variables. This example is typical of many, and is given in detail. (The beginner may also like to solve it by straightforward partial differentiations.) We assume throughout that the context makes it clear when ¢ is being regarded as a function of «, y and when itis being regarded as a function of u, v. By the chain rule of §10 (p. 25) 26 _ apeu , hav Bu Be * Ov Oe =F (20+ Fay; % A$ Ou , 26 20 = bu By * bv by and = 8 (—2y) +8 22), Hence ery fe = mat) + say 2b = uF 4 228, Wat, # say anit 26. a aey 26 = 205 Fy" and 4244 PARTIAL DIFFERENTIATION We may regard the symbol ttle, as an oPmnaTor in the sense that, when it ‘operates’ on the function ¢ (expressed in terms of x, y), the aay is 2 we ty, vis. ‘What we have just proved is that the operators Brien Dpesit ae Yay? V3q— iy? acting on any function He a, ct are equivalent to the operators a a mud +202 ae 25 ay operating on that fale ae expressed in terms of u, [The next steps may be abbreviated with experience, but are given in full for clarity.] In particular, a 4) (2 2 a) ty x) (oh, (ar9g, Yay songs Lifegits eg = (54202) (2u58 +2086), since the functions oh, yt, aust 4 2988 and the operators Boned Bvacid : ' (‘a+¥a) i (+ 202) are respectively equivalent, Consider the left-hand side. [This step should be noted carefully, as it is often osloulated wrongly.] This is equal to (dong) CB) -bend) 8) ILLUSTRATION: USE OF OPERATORS 45 saat) +42 ro) w(t bh B) bray Ae + sp ereglays, se we ee the equation ae Saye 2 ot agit = 4(wih +2 28 ee. bo Be tesere), Now consider similarly the relation a) VE-*%) = (5-242) (2088 os ~2ui8), The left-hand side is rate)46 PARTIAL DIFFERENTIATION Hence wes tay OE tO gt 2g Seay” ay? en Yay 422? — ony 2b 4 OP _ 4,28 _ 208 “(5 2ue=—— auto t™ dot - du ae) Adding this to the similar equation obtained above, we have (@ +(Ee eer +53) = (ut +0) (e+ 6 +8). But wttet= (ty) ay)? = (tev L 2) _ (08,8 so that ainleetae) -4 +58). si son 28 28 : srsmartox 12, The expression £8.45 in polar eoordinaes jaw e expression een tet tye is of great importance in mathematical physics, and it is important to be able to transform it to polar coordinates in accordance with the usual formulae z=rcos8, y=rsind. We use the method of the preceding illustration, now expressing the argument with more normal brevity. By the chain rule of §10 (p. 25), so that tae Hence rZ (rt) = (eZ+yz) («se +uz8), oe 8 0b ty OE pe ce 4 yt, or HOMOGENEOUS FUNOTIONS 47 ie 2 _ OB 2e,, Wey 20 ~ bx 00 dy 00 =F (—rsinay + $2 0080) == yb sett, gray tg ltt call ag ®t. Adding, FE SE OE ty (Beret) = ae ze), oat * Bye an a6 126, 10% oo Se Set aoe InnustRation 14, Homogeneous functions. A function of three variables f(x, y,z) is said to be homogeneous of degree n if it possesses fhe property ua, uyyw2) = utftes2) for some number n. Typical examples are a+ yt ey te eas (n= 0), 1 a + 2y8+ 328 Another example is yz+ /y+/z (n= 4), where it is assumed (for real functions) that x, y, 2 are positive, and that w also is positive. We prove the following theorem: If S(e.y,2) isa function homogeneous of degree n in x, y, z, then is Fat Vay Suppose that z, y, z are temporarily fixed, Then (ua, wy, ua) (n=), (n=-3). 42 Lang,48 PARTIAL DIFFERENTIATION is a function of the single variable u. Write uc=E, uy=9, w=. Then, in accordance with the rule for differentiation (compare Illustration 5 on p. 24), a Gal (ue ty, uz) = Ley od Fen Oe Vere du" du ar KEnO ty AE 6) , Ene) ogy ras ope But, by nak Sux, uy, uz) =u"f(e,y, 2), so that (2, y,2 still being fixed) we also have the relation © jie, ny, 2) = 0-0, 9,2). Hence, equating values of 4 f(ue, wy, u2), elms) , FE) , Ene) Pane atte oy OER In particular, this result is true for w= 1; and then &, 7, { are respectively equal to «, y 2. Hence aaa +y =nu"-¥f(x, y,2). olen) ue y Lee) ee or, more briefly, LryL aL ont This result may be extended. By differentiating with respect to z, y, 2 respectively, we obtain the relations of of ef ef of "iat bet Yaeoy + *dzon = "Be" ay x, ae "aay t oat ayae~ "ay" yf, ee eee TY ayaa tae Oe = be" IMPLICIT FUNCTIONS OF A SINGLE VARIABLE 49 Multiply by 2, y, 2 in tun and add, first transferring the terms in te # to the right-hand side. Then Ga’ dy’ Oz ey hae nL + one es oc — oo feLeye o-Q) =n(n—1f. 17. Implicit functions of a single variable. A function y of a variable « is often defined, not directly in the form y = f(z), but implicitly by means of an equation Slw.y) = 0. ‘The differential coefficients My #y of y with respect to « may then be calculated as follows: : Since x, y are both functions of 2, we have, by the chain rule, the relation Ut ey dz ~ ta" yd’ so that, since f is always zero, O= vs Hence Be af af Moreover, applying the same treatment to 2°, 2) respectively, seca ee de \Bx) ~ Gn * Gedy de’ d (af\_ ef ef dy a (by) ~ ty ra so that, on differentiating the relation ont ty ae ay de50 PARTIAL DIFFERENTIATION with respect to 2, of, dy) (OF Ma dy Yay _ 4 G2 Seay dex) * \Zedy* dy? dx) dz * dy da? ~ a 2 of wy ef “ of dy S “saci * ayaa = > af af of Os Lie Intusrration 15. Zo find expressions for Ht wey ahem @+y—1l=0, Differentiating, we have a+ yt = =0, so that ie, Differentiating again, EXAMPLES XI in each of the following cases: 2, aa?+ fy*—1 = 0. 4. ayttyat = 1, 18. The tangent to the curve f(x,y)=0. Let (p,q) be a given point lying on the curve whose equation is S(e,y) = 0. y by . de 1. y—4ax+a? = 0. 3. a +y? =a. Find expressions for z TANGENT TO THE CURVE f(x, y)=0 61 ‘The gradiont at the point is, where a a dy & a” F ey (p. 49), the differential coefficients being evaluated when z= p y = q. Hence the equation of the tangent is iB wr or P)» ol [Een oe said the square brackets being inserted to imply evaluation with r=p,y=q Innusrration 16. To find the equation of the tangent to the ellipse 2 =4+%-=10 at the point (p,q). pala tte We have or Wie af at pcan [ie artim oe a Be Hence the equation of the tangent is Pep) + ya =0, or, since pHa geet = 1, Fa a1, 19, Exact differentials. We are now familiar with the formula dz= Ede & dy for the differential of a function z of two variables 2, y; but the converse problem often arises and requires an answer: Lo determine whether, when two functions f(x,y), g(x,y) are given, there exists a function z of x, y whose differential is given by the relation dz = f(x,y)de+g(x,y) dy.52 PARTIAL DIFFERENTIATION If z does exist, then, necessarily, E-few, 2-0, and these two partial differential coefficients satisfy (for normal conditions) the relation (p. 38) i) ap @ (ez a by fe) “be Hence a wecrssary condition for the existence of the function zis that a @ ay Sm} = a, (ole w))- The condition is also surFiormeNt: If two functions f=f(«,y), 9=9(«,y) are such that 7 ey ea? then there exists a function z of z, y such that dz = fdx+gdy. [This proof may be omitted at a first reading.] Take the function f(z, y) and integrate it on the assumption that y is to be kept constant; this gives a function Few=[ Seaunde aan In the same way, integration of g(z,y) on the assumption that 2 is to be kept constant gives the function Ga,y) -{ o(z,y)dy. (e const.) ‘The functions F’, G satisfy the relations (Vol. I, pp. 86-7) oF wot a wy a so that, in virtue of the given relation zg = Es we have ar _ oe Wye” aay EXACT DIFFERENTIAL 53 If we assume that F is a function of ‘normal’ type for which oF _ @F ayen Gxey’ ar _ ee Gady ~ Gaey’ ar-@) or exdy a Hence (compare pp. 42-3) the difference F—@ is of the form F-Gsu(z)-vy), where u(z), v(y) are functions of «, y only. It follows that F(x,y)+ vy) =G(a,y) + ule), and we take either of these expressions as the function 2 whose differential we require, namely, z= F (x,y) + vy) =G(@,y) + ula). With this choice of z, we have a _ oF oe Ge a _ 2 wy oy =o a, that ‘de +gdy =~ da+—~ dj 80 Sdorgdy=sdets dy 0. = dz, Drrmaition. When the condition Zen) = Ztotem is satisfied, the expression Sle,y)da+ g(x,y) dy is called an EXACT DIFFERENTIAL. Inzusrratioy 17. The steps of the proof may be grasped more clearly by reference to a particular example, say Ba%(a+y) da + (z+ 3y) dy. Integrating the first part with respect to x, on the assumption that y is constant, we have F(z,y)= jet +2%y,54 PARTIAL DIFFERENTIATION and integrating the second part with respect to y, on the assumption that zis constant, we have Oc, y) = ay + hy? What, in essence, we have to do is to add to F(x,y) a function of y, and to G(x,y) a function of «, so as to yield (if possible) same function z. This we do by taking ee fottaty + iy Then dz = (323+ 3x%y) de-+ (a + 8y) dy, as required. The point of the proof given above is that, since 8 a ag" +30) = 5, (Ot + Baty), functions u(x), o(y) must exist such that F(w,y) + vy) =G(w,y) + u(x). ‘When they have been determined, either side of this identity gives a function 2. 20. Integrating factor. Although there is not, in general, a function z whose differential dz is equal to the given expression fdz+gdy, it nevertheless happens in several important instances that a function 2=Ha,y) exists with the property that the product MM fda -+gdy) is an exact differential, The function (x,y) is then called an INTEGRATING FACTOR of the given expression. The condition given in the preceding paragraph becomes e é Sled) = gaia), y 8, o or wl ash = nro, a or INTEGRATING FACTOR 55 ‘This equation, which we may expect to be of help in deriving 1, is unfortunately somewhat awkward to handle; but in practice what we often want is a solution in which is a function of x only, in which case the working becomes more manageable: If is a function of z only, then On ae ey and # is tho ordinary differential coefficient Ps Then ae (te ne a(t 2), Ff _ so that ceiaie ar] If it happens that the coefficient of dz is a function of « only, the function u(x) can be found; for then ‘The Illustration which follows shows how an integrating factor may be found—though an experienced mathematician will usually ‘spot’ the solution in such a direct example without going through the routine process. Inustrarton 18. To find the current y at time t in an electric circuit with self-inductance L and resistance R under an electro- motive force Ecos pt, where L, R, B, p are constants. It is known that y satisfies the differential equation dy Lat ky = Ecos pt, or, on multiplying by the differential dt, Lay +Rydt = Boos ptdt. We seek to find whether the left-hand side is an exact differential. If so, 5 is ql) = 5, (Rob or o=2, which is not true in general.56 PARTIAL DIFFERENTIATION We next try to find an integrating factor 1, by means of which’ the equation will become pLdy +pRydt = nll cos ptdt. ‘The right-hand side can be integrated (perhaps with difficulty, of course) if is a function of t only. Let us therefore try to make the left-hand side an exact differential on that assumption, We must have a qe) = Zuko, or, since # is a function of t only, ae La aer, dn _R or aL Hence log = Ri/L, or paeMz, no constant being introduced since any one particular value of will suffice, ‘The equation is therefore LeMLdy + Rye™Zdt = HeM“ cos ptdt, or Ldly eB) = Fe®% eos ptdt, or yell = G/L) [e% cos pet o, where C is an arbitrary constant. Evaluating the integral by parts in the usual way, we easily reach the solution yemt Becaa Reospt+ Lpsinp} +0. The current y at time t is therefore given by the formula E y= Bip rapa Boos t+ Ipeing} + Oem, where is a constant whose value, for a given problem, depends on. the initial conditions, TAYLOR'S THEOREM oT 21. Taylor's theorem for several variables. In Vol. II, p. 44, we discussed Taylor’s theorem for functions of a single variable. In enunciating the extension we have to use two points of notation which we shall explain almost immediately: ‘Tavton’s rHeoneM: To prove that, if f(y, 2) isa function of the three variables «x, y, z, then a number 8, lying between 0, 1, can be found such that Seth ytj2+h) =fey.2) +(rZnig+rg) tema +a (tgrigeeg) fe 4,2) ak Vig inthe)” te 2) Fea hae tay tee, a a,,2 ey It is assumed that the function and all its partial differential co- efficients exist and are continuous for the ranges of values to be considered. [For the ‘shape’ of this expression, note the analogy with the binomial theorem.] FEZ) f+ 0h, y 405,240) Noration. By the expression (3 +9 igths) S(@,y,2) we mean the result of operating p times in succession on the function Sle,y,2) by the operators tin +k Ss foeeauple whaap = ies we have respectively LiLo ef, wh paws Sh at E+ aun 2+ mye58 PARTIAL DIFFERENTIATION In the term (« Btigtkgy ‘{le+0h,y+0j,2+6k), it is understood that, arrmr the n operations, the variables x, y, 2 are replaced by «+ 0h, y+0j,2+0k. Suppose that «, y, z have definitely assigned values. Form the function F(t) of the variable ¢ by means of the relation F()=fe+th,yttj,2+th). If, for convenience, we write u=ax+th, yttj, w=2+th, then (2, y, z being temporarily constant) = A du , Y do , of dw = bu dt * Bvdi * bw dt ant Lad, (Compare also p. 48.) Thus the operation da a on the function F(t) is equivalent to the operation anyeliea ha tlaet hag on the fanotion f(u,9,1). ee a function of t, and to its alternative expression wf +e, which is a function of u, v, w, we obtain the ete @ (,8,,8,,a)\3 aan (ba tig the) . and so on. Now, by Maclaurin’s theorem (Vol. IL, p. 49), 8 exists (0<0< 1) such that F(t) = F(0) +tF"(0) +... : FO-0(0) +5 FOO, ee t@=— TAYLOR'S THEOREM 69 and so, using the relations just obtained, f(t, v, 2) =f(u,v,w)y By Oh ae +i(ncrigred) lu, v,10)y mapa 8 a\et tean(tatigtigg) Nuch +5 (trigthe)’ Mee le where, on the right-hand side, the suffix 0 indicates replacement of t by zero in f and its differential coefficients, while the suffix Ot indicates replacement by Ot, APTHR DIFFERENTIATION in all cases; thus 2 Lt (stig tke) foemms a 2 a means BA feane) 45g flew) the slenth Enea and (igrad+eg) 110, 0,t0)m means (42452 422)" fees not,y +0h,2+-b00), a thayt*s, ny +90 2+ KOE), the values +01, y+j0t, 2+KOt being inserted in the partial differential coefficients arrun differentiation by the current vari- shlew ee Putting ¢ = 1, we obtain the theorem. Tzusreamton 19. Taylor's theorem for the function aye. Tf S(@,y,2)=2y2, aoe . then (Stig tke) tena = hyrtjort hey and (sgeigeeg flee. y2) = Spe + hy + 2je, oe Jy Be) Ys %) = 25 y60 PARTIAL DIFFERENTIATION Hence () anumber @ exists such that (@+h) (y+j)@+h) = ay: thyz+ jew kery + H{2jk(w + Oh) + 2kh(y + Oj) + 2hj(z+ Ok}, so that, on expanding and cancelling like terms from each side, jk = Bhjkd, and O=3; (i) a number ¢ exists such that (ath) (y+j)(@+h) = xyz + hly +99) (2+ Pk) +j(2+ Pk) (w+ Ph) +k(c+ gh) (y+ 4), so that, on expanding and cancelling like terms from each side, jke+ khy + hjz-+hjk = 29 (jhe + khy +hjz) +3¢%yjk. ‘Taking the general case, in which h, j, k are not zero, divide by hjk; then ag (2-1) (FM Itis an interesting exercise in elementary algebra, which we leave to the reader, to prove that, if the expression 2y.t ce does not lie between —1, —2, then there is precisely one value of between 0, 1; but that, if the expression does lie between —1, —2, then there are two such values of ¢. In either case, the theorem is verified. REVISION EXAMPLES X “Scholarship” Level 1, Prove that, ifz is a function of w, », where usa, v=rty (60 that z may also be ae as a function of x, y), then a ae fan tat Neate ae REVISION EXAMPLES X 61 2. If the independent variables z, y in the function f(x, y) are changed to £, 7, where g=aty, 7=2-y, so that the function f(x,y) becomes g(£,7), prove that Deduee, or prove otherwise, that, if the independent variables z, y are changed to u, v, where ety =log(ut+v), 2—y=log(u—v), so that the function f(x,y) becomes h(u,v), then 3. (i) Prove that, in general, ( ech (Z-»g) f(e,y) ). " ey is not equal to eet matt ay (F- f (ii) Prove that, if is a function of u and u=(@+y)ton2, then of -yZ - wre wt (herd) (v8) nearer 4, If caw, y=uto, ae prove that heen If ¢ is a function of 2, y which setisten the equation === = 0, prove that Cd a2) oz) (ume) (SS +053) — 08 oe — wt n)(Z -@) -0. [A permissible method of proving the second part of the question is to assume that z is of the form f(z) +9(y).]62 PARTIAL DIFFERENTIATION 5. If g=ay, y=xt+y, See ere Prove that aee= tan aaa e 6. u = e*(voosy—ysiny), prove that fens o=0, a (a+ 58) - erty. If uw=e(veosy—ysiny), v= et (siny+y cosy), roman eee dea annamiumptag el en rita tcc 7, Variables z, y are defined in terms of u, v by the equations z=ccoshucosv, y=csinhusine, Write down the values of I eu’ Ou’ dv’ Ge’ and prove that Remtegeesr = 4e(z+y) sinh u cose, Zeru +5 Zorn = 4c%(cosh?u—cos*v), 8. The function u(x, y) is defined by the formula w= e(xcosy—ysiny). Find a function v(@, y) satisfying the conditions ae ee oes Oe ety’ ty ee? | (0,0) = 0. Show that u+4v is expressible in the form F(2), with 2 = 2-+iy, and find the function F. REVISION EXAMPLES X 9. (i) Prove that the equation ay or ey, oF seta terran o z atyt+z is satisfied by i=s59a (ii) Given that @ =e" cos2uv, y = e-*sin Quo, ae _ty ty & prove that Bde ee Prove that, if fis a function of 2, y, then H= ue—- Lr 260 +00), z. = aay +00) L 4 2a) Henee prove that ( is the same for all functions f. 10. Given that z=cooshucose, y=csinhusine, where c is constant, find expressions for oy Gu’? 20" Given that f is a function of u, 2, prove that 2 iL a esinh (u— ~in (Le 2), ey)? 2% eteaita—catn (2) Gu! u v UL, Given that 2-555, y=ao, and that fis a function of x, y, show that wry (Sh Cells +54) - way (Fre .64 PARTIAL DIFFERENTIATION 12. (i) Given that uaety, vea+y’, prove that, if « is considered as a function of u, v, et eS 2x(e—y)’ 0 Bx(@—y)" Prove that a ee ae Be q ay ey Gxy(z—y)° bu a0 (ii) The variables @, y, @ are connected by the implicit relation 22+ y%— 2ay cos = sin®d. Prove that @ = + (sin-1z—sin-1y), and deduce that ma 13. If Se SRE — cond +-conh, Ge gee peg ise B78 37 2 ‘The function u(z, y) transforms by means of the above relations into (0,4). Prove that Sahat. ae (=) +(63) (é ey) ~ (cosO+ cosh )* 14, The identical relation fe 23, w2—y8,u8—2) = 0 defines u as a function of x,y,z. Prove that 10m 100 Lou _ Le aaa ydy "2 oe 15, Prove that, if z=rsind cosf, y=rsinOsing, z= rcoso, and ifr is a given function of 0, , so that z may be regarded as a function of 2, y, then (Fsint 0+rsin 0080) & + Zin p-+rsin*Ocos —% sind e030 cos =0. REVISION EXAMPLES X 65 16. Prove that, if e=utetw, y=votwutu, z= uw, and if z is a function of z,y so that w is a function of u,v, then tog a eu ss ee = ww—p—Gu+w) wu—p—qw+u) a a where e-em IR 17. Show that, if 2, y are functions of £,7, and if «= () +): ER ae =) +) then, for any differentiable function U(x, y), QU _eU ey Woy ee ob Oy Oy ee OU ex OU te Boy 7a oy oy Prove also that 400 we 2 aU eu 12 oe en Gat Bt HE — He 18. A function f(x,y), when expressed in terms of the new variables u,v defined by the equations w= dute), Yow, becomes g(u,v). Prove that og =e 2 OF OF Budo ~ 4\eat Ty Bucy” Oy?” yey)” 19. Prove that, if f(z,y) is a function of 2*+-y* only, it satisfies the identical relation i= ot "Oy By changing to polar nee or otherwise, prove con- versely that, if/(,y) satisfies this relation identically, then it must be a function of 2*-+-y? only.66 PARTIAL DIFFERENTIATION 20. Prove that, if 2, y,z are functions of two variables u,v given by the relations zafue), y=gue), 2=hu,2), then (suffixes denoting partial differentiation) Cate) = tude pu, | ft Se Sead Par dude + food (Ludo fodu) 5 =| Ju Go IuuIs~ 2DuvIuIot Ive | Ia Ihe PauSt- Aru Gudot hoes 21. The function y of two variables 0, is defined implicitly by the equation y= 0+29(y). If wis a function of y and F(u) a function of w, prove that alt |= 2 (rong, andthat = Su)iy, Er gp CUS | and, shi that St ga our. 22. Prove that, if z=rcosd, y=rsind, and if ¢ is any function of r, 0, then years and obtain a corresponding expression for . Prove that, if g=r*sinnd, then os +58 Ws) 23. Prove that, if U=f(ely), U,=10, ew where r? = 2+, then eB tyy ~% 2U, 2U, U0 eu nt attest arte?) REVISION EXAMPLES X 67 Hence, or otherwise, prove that, if the equation au, Ou ee Sy a aye is satisfied by u = V(x,y), where V(z,y) is a homogeneous function of degree n, then the equation is satisfied also by u = r-*"V(z,y). 24. Four variables ti, tg, ts, uy are such that any three are independent and each may be expressed as a function of the other three, Tf aye denotes 34, the rate of change of u, with respect to tl, when tg, u, ate kept fixed, and if u,, (i+j) has a similar meaning, show that Upstsithe =—1, thes, = Ura aa 25, The variables 2, y, z satisfy the two equations F(e,y,2) = 0, g(ey,2) = 0. By eliminating z between these equations, it is possible to obtain a relation connecting z, y which defines y as a function of z. Show that dy _ _f2Is-S9e de fyte— LaDy" 26, Given that 2, y are functions of u, » defined by Fy, 4,0) =0, g(z,y,u,0) = rind = in terms of partial derivatives of f, g with respect to z, yw, v- If a+y%—25uv = 0, ue-+vy—1=0, prove that “© = 4 when u =o = 1, and give the reason for the ambiguity in sign. 27. If f(x,y) is a function of z, y, where 2, y are functions of t defined by the relations u(x,t) =0, v(y,t) = 0, and if f(x,y), when expressed as a function of t, takes the form 9(¢), prove that dg__faf wou pee Bu Gv at ~~ \exdy Gt * by Oa ot|/ Gedy"68 PARTIAL DIFFERENTIATION 28, (i) If w= xyz, where , y, z are connected by the relations yeteetay =a, x+y+2=56 (a,b constants), prove that es = (%—-y) (wz). (ii) If, 7 are functions of «, y such that §=ecosy, 7=e*siny, and if z, y are functions of r, such that a= ecos0, y =ersind, where r is a function of 8, prove that ae a By F tan(y+0) 29, The equation z= F(x,y) is obtained by eliminating u between the equations y=f(ux), 2=9(u,2). Prove that for Ya Ye 30, Prove that, if wis function of x, y which is transformed into a function of r, 0 by the relations 2=rhcos}0, y= risin}6, ay op _ Ou, | Ou then egy @ we aa tY ay seo Gi) 25 = ain Ou | Ou it) = Bu, Ou (ii) a(t +e tia = mt ae REVISION EXAMPLES X 69 $1. Given that 2=reos0, y=rsind, 20 20 80 find a aay and prove that 20 =n aaron. ancy 32. oe ee cee zeos0—ysin@ = zsin+ycos0 =r, 30 a 80 Or find Be ae’ ayy in terms of r, 4. Show that, if V is a function of z, y, ev ev 1feV 1eV, 1a laos 33, The variables 2, y are changed to r,0, where w=rsec0, y=rtand. Show that Ou Ou du 1du_cos*#du _ sin# cosd du te dy oA ror ae = 2: 84. Given that u,v are functions of the two variables x,y, @ expreas| & S +m) (u»)in terms of the partial differential coefficients of wand v. If wis a function, with third-order differential coefficients, satis- fying the relation =" +2 = 0, and if Ga? * Oy w= (e+y*+a0+ byte) u, show that (St@) e=e. bat * Syl 36, Four variables 2, y, r, @ are connected by the two relations x=reosd, y=rsind. Asymbol like Z) indicates that ris exprosod asa function of and that y is kept constant during the differentiation of r with respect to z. Verify the relations « @),-@) Gr} \dx),” © @&),@),-" (0 --@, G,-@, =e70 PARTIAL DIFFERENTIATION Illustrate (iii), (iv) geometrically, taking (x,y), (r,) as Cartesian and polar coordinates of a point in a plane. 36. If A is the area of a triangle ABC, calculate a when the independent variables are @ 4,6,0; (ii) a, B,C; (iii) a,6,0. Illustrate your results geometrically. 37. If the circumradius R, and the area A, of a triangle ABO are regarded as functions of 6, c, A, prove that aReR a Sr eas ee aReA aReA i et ae op 7 FRSA. 38. If (x,y) are the Cartesian coordinates of a point and (r,6) its polar coordinates, define what you mean by =, a Find their values, and illustrate by a diagram, Prove that es = ( a. 39, Ifw =f(r), where w= rcos0, y=rsind, Ou ou _ df df poi Bat Gyt — dr? ede” 40. If z=eflylz), ue Logon eat yee ees =r(r—1)z. 41, If = flan? + Qhay + by?), 0 = glax*+ hay + by*), rows 58) -2(8) BRVISION EXAMPLES X m 42, A function f of two independent variables r,6 is trans- formed into a function g of variables u,s by means of the relations roos@ = 1ju, tand =<. Prove that Lau, Faw % 0409, Prove also that, if f satiafies the equation of 0030 L +rsin of = 0, then Boaseygeo, where ¢ is some function of u. 43, Ifz,y are the coordinates of any point and a=rcos0, y=rsind, find the values of 5) cna’ ee Verify the second result geometrically. Prove that a +a 0. 44, The variables 2, y, z are connected by the relation. ty tet Bays = 0, and P(e, y, 2) =aPye, Determine the value of a at (1, 1,1), (i) when the independent variables are x,y, (ii) when they are z,z, and explain geometrically the difference between the meanings of a ae 45, Each of w, v, w is a function of the three variables 2, y, 2. Whenever «, y receive small increments which satisfy the equation72 PARTIAL DIFFERENTIATION dy—zdx = 0, the corresponding increments du, dv satisfy the equation dv—wdu = 0, Prove that av_ au, (av a) St (2 oS) mio, 2 werte(e “a 2% =o, Find the most general functions u, v, w for which w is a function of z only, v of y only, and w of z only. 46, If $(w,y) is a differentiable function of z, y, and if sinw ake Se) ~cosuoosho’ Y~ cosutcosho’ show that, for a certain form of the function f(v), 2 28 og wae tay Be and find the form of f(v) in this case. 2 26 = sinucosh v= +-cosusinh v=, 47. Ifw=a/y, v = xy, prove that Mail - oe sot sn2), 48. The area A of a triangle is found from measurements of the side a and the angles B, C. Prove that the error 6A in the calculated value of the area due to small errors da, 8B, 8C is given approxi masely by dA_,8a,¢ 3B b 80 Bn a tasnB and" 49. The side b of a triangle ABC is calculated from measure- ments of the side a and the angles B, C. There may be an error not exceeding h (in either direction) in the measurement of a and an error not exceeding « in either or both of B, C, where h, « are small. Show that, if B+C <}m, the greatest error in b is approximately oftracora}. a Find the greatest possible error in b when B+-C> 7, and explain the difference in the two cases. REVISION EXAMPLES X 3 60. If S(x,y) =axy+be+cy+d, g(x, y)=a’ayt+b'a+c'yt+d', @ Ply) prove that aff 2, where p(y) is a quadratic function of y (not containing 2). Hence or otherwise prove that f(x, y)/9(e,y) is a function of the product XY, where X is an appropriate function of x only and Y a function of y only. 51. A function y of 2 is defined by the equation f(z,y) = 0. dy Py, Express $Y, 5-4 in terms of p, 9,7, 8, t, where 7 “3 ots gual oy ex’ oy If the same equation is regarded as defining « as a function of y, prove that = pe = gt. ay 7 ae 52. Show that, if « = peos¢, y = psing, then AE Roca (2 ov ee ee ap Op pep OF Hence show that gate aleve z 2 (p+ 2/4 tan*% are all solutions of the equation av av ev _ wet yet ae n>74 CHAPTER XV MAXIMA AND MINIMA 1, The general conditions. Suppose that u=f(e,y,2) is a function of the three variables z, y, (not necessarily indepen- dent). The function w is said to have a maximvat at a point P, for which =a, y =b, z=, if its value at P exceeds that at neighbouring points, and a smxmrum if the value is less. As for functions of one variable, maxima and minima are local properties. ‘The argument for several variables is naturally more complicated than for one only, and the treatment which follows is of necessity more sketchy. We begin by finding necessary conditions, assuming first that the three variables x, y, 2 are independent. Suppose that the function usf(x,y,2) has a maximum value when # =a, y = 6, 2 =. Its value is then greater than that at any near point; in particular, at all those near points for which y = 6, z = c, Thus the function f(@,b,0) of the single variable z is greater for « = a than for any other near value, so that this function has a maximum atx = a. Its differential coefficient with respect to x (while y, z remain constant at b, c respectively) is therefore zero when z = a. Hence af Leo when 2 = a, y=, z=. Similarly ¥ 9 Fao oy ae for those values, ‘The presence of a minimum may also be treated in the same way. Hence a set of necessary conditions for the function w=f(r,y,2) (x,y,z independent) THE GENERAL CONDITIONS 15 to have a macimum or a minimum value for x = a, y = 6, = cis that F oc o es Br ay tes. fora =a,y=b,2=0. Since these conditions must hold at each maximum or minimum, they enable us to locate such turning points. But the conditions are not sufficient; moreover, they do not, of course, distinguish between maxima and minima, Detailed analysis is beyond the scope of this book, but common sense will often provide the answer for particular problems. ‘The three conditions may be gathered into a single differential form. For the differential of the function w is ata tan’, uma dat ag tet aos which vanishes at a maximum or minimum. Hence the differential dun aes Lays Las vanishes when w has a maximum or mini- mum value, for arbitrary values of the differentials de, dy, dz. Expressed in this form, the condition du = 0 for a maximum ora minimum may be applied whether «, y, 2 are independent or not. For if they are not independent, the conditions of dependence enable us to reduce the number of variables until those which remain are independent, and the condition du = 0 then follows. Note. If «, y, z are not independent, the condition du=0 for a maximum or minimum is still equivalent to ou Ou Ou ae gta 0, but it does xor follow now that 24, a vanish separately. For the treatment to be adopted, sce §3. Innustration 1. To find thegreatest distance of a point onthe surface ax? + by? +02" + 2fye + 2gee + hay = 1 (assumed to be an ‘ellipsoid’) from the plane z = 0. (The surface may be pictured as a somewhat distorted sphere with its centre at the origin.) es Ry 276 MAXIMA AND MINIMA TO DISTINGUISH BETWEEN MAXIMA AND MINIMA 177 The distance of the point (z,y,2) on the surface from the plane z= 0 is simply the coordinate z, and, at a point of maximum (or 2.* To distinguish between maxima and minima. Though as il ae : venta 02 08 detailed consideration would us much too far, the general minit distance the two partial differential coeffi a eee) i ae ee ate cients oo Oy picture may be exhibited of a method to distinguish systematically a eee 5 4 between maxima and minima, Differentiate the equation ofthe surface partially with respect to For, say, three independent variables, suppose that the function «, and then put = = 0. Thus Welegn ax+thy +gz = 0. has a turning value at the point (a,,c). It will be a convenience Similarly, he + by +fe = 0. of notation to express the conditions in the form These two equations, together with the equation of the surface, Bu determine the points of greatest distance; the z coordinates give fare ob a0 the actual distances. , ialeetiee toi inil ya aliesate 29 aa follonn If £, 7, € are small, the value of the function at the near point Multiply the two equations of condition by 2 y respectively, and +E btnet+) 8 ai eos meth), subtract from the equation of the surface; divide the resulting 5 Seetalty chee : where, by Taylor’s theorem (p. 57), get fy tee = 0. fart, b+y,0+6)-f(a,b,e) Eliminating x:y:1 determinantally between the three linear = (gengreg)seun equations, wehave | q gz EY 8 1/,2,, 9,0) bb fe |a0, +3 (at1ytSs) f(e.y,2) 1 Li tae tees or, after division by z, the differential coefficients being evaluated at x =a, y=b,2=c. aeiRigh emo Taking , », ¢ to be so small that powers and products of degree greater than 2 may be neglected, and remembering that bosib 6 0 |g [=o OF OM NON 5 aes, Expanding, we obtain a relation which we may write in the form (abe-+ 2fgh—af*—bg2— cht) — 3 (ab—It) =0, Ha+£,b+m,e+0)-f(a,b,e) ab— 18 aCe Bie Brose ge acres e425 o, so that z= +f Gerace 3\eai® + aa” tact® **abe0* “dean? * * Gach From the shape of the surface, this gives the greatest distance, where £, 7, (are variables, depending on the point selected near to whose value is therefore (a,b,c), and where the coefficients Fu. are constants, ab—ht Jemma aon) '* This paragraph may be postponed, if desired.78 MAXIMA AND MINIMA UNDETERMINED MULTIPLIERS 79 For a maxruuat this expression must be negative for all sets of Thus the function f(c,y) of two independent variables has a values of £, 9, ¢; for a muNIMUM it must be positive. This condition maximum at the point (a,b) if is both necessary and sufficient. If, however, the expression varies a a in sign, being positive for some values of £, 7, ¢ and negative for an! others, the function u has neither maximum nor minimum. es ose.) ‘The simpler cases of genuine maxima or minima can be settled ea* by definite algebraic tests: HOY _(B)*, o, (i) For convenience, write the expression with the notation Ga? Gb (eh) Es Al + By? + O68 + 28 yf + 20¢E + 2HEn, f of of 5 and a wesmiom if =0, Z=0, and form the determinant whose rows are the coefficients of 2£, 29, 2 in the partial differential coefficient ta ee Sp spectively, 0, namel AHG ly» en ()> > 0. HB Fi. aa®eb? \da db) GFe (Note the direction of the last inequality.) Ee sara et tne ee ete a ane (ii) The method can be extended, in an obvious way, to any emus isha ee hh a ASNT ta calc. number of variables, For the expression sf ‘ ie AE 4 By OC + Dre 2B + 265 + 2HEy+ 2UEr + BV 9+ Er, the determinants are ‘Then it is known that a necessary and sufficient condition for E to be always positive, for all values of E, 9, & is that these three deter 4 2 IC Lol | peed wl tA th minants should all be positive, Herv|,|o 8 F\, |e B Consider, for example, the expression orow|lere BE? + 27? + 26? + 26 — 406 — 4Ey, GV WD The determinants are 3-2 -2 3. Lagrange’s method of undetermined multipliers. The Aglats. rien method now to be explained may be grasped more readily if we eas begin with an illustration: een Intwsrratton 2.* To find the lengths of the axes of the conicin which bigs at the quadric qu0 by2-4 c22-+ Ofye + 2gen-+ Qhay = 1 and ei: meets the plane le+my+nz = 0. Since they are all positive, the expression is positive for all £, 7, £. heen cio ok asi th Sheory of ee ela aay Oy ii ive i Teg ‘the problem as the determination maxi minimum values n co eo ee for E to be always negative is that —H should be ne ei Pesta is yasectlon ey 9 eens ie ‘equations quoted in the enunciation.80 MAXIMA AND MINIMA The squares of the lengths of the axes are the maxima and minima Bee auaoa wae +ytte, At such points, du = 0, so that ada+ydy+2dz = 0, Since the point (z,y, 2) lies on the quadric, the differentials satisfy the equation (ax+hy +gz) da + (ha-+ by +fz) dy + (gx +-fy +cz)dz = 0; and, since it lies on the plane, Idx +mdy+ndz = 0. Multiply these three equations among the differentials by 1,0, respectively, and add. Then choose A, 4 80 that the coefficients of dx, dy both vanish; that is, so that a+A(ax+hy+gz)+ul = 0, Yy tA(hat by +fz)+ pm = 0. Then [z+A(gu+fy +cz) + pm] dz = 0. Now the three variables , y, z are subject to two conditions, namely, the equation of the quadric and the equation of the plane; their three differentials are therefore subject to two linear con- ditions, so that one differential may be given an arbitrarily assigned value. In particular, dz may be assigned arbitrarily—it need not bezero. Hence the coefficient of dz in the equation last written must besero, so thet Age +-fy+es)-+ am <0: To summarize, the five equations aa? + by® + cz* + 2fyz+ 2gzx+ 2hay = 1, le-+-my +nz = 0, a+A(ax+hy+gz)+pl = 0, YytA(ha + by +fz) +m = 0, z+A(getfytcz)+un =0 serve to determine the five unknowns consisting of (i) the three variables «, y, 2, evaluated at a turning point, (ii) the two multipliers A, #. The sacrifice involved in increasing the number of unknowns is amply repaid in symmetry. UNDETERMINED MULTIPLIERS 81 In problems of this type, where the equations of condition are homogeneous polynomials in x, y, z, a good first move is to multiply the equations involving A, u by 2, y, z respectively, and then to add them, This gives @+y? +2) + A(ax* + by* + c2*+ 2fye+ Qgen + Zhary) + p(la+my +nz) = 0, so that w+A(1)+p(0) = 0, or u+aA=0, or Aaa Hence w—u(ax+hy+gz) +l = 0, or (-f)2+v+oe—4r 0. Similarly, tart (6-2) vse = °, ort fy (eZ) 2~fn= 0. Also la-+-my+nz = 0. Eliminate the ratios : Then jw between these four equations. ao we igh 7 , ine gf on Te wo Thisis, on expansion, a quadratic equation in >, and therefore in, whose roots determine the maximum and minimum values of u. The lengths of the axes may thus be found. The next illustration will help to consolidate what has been done, and also to expand certain details.82 MAXIMA AND MINIMA Innusrration 3. To find the maxima and minima, for NON-ZERO values of x, y, 2, of the function usazyz, under the condition atyte=1. At a turning value, du = 0, so that yude + zxdy +xydz = 0. By the equation of condition, da+dy+dz = 0, ‘Multiply these two equations among the differentials by 1, A respectively, and add. Then choose A so that the coefficient of da vanishes; that is, so that 4, 4 4 _ 9, ‘Then [ex+A]dy+[zy+A]dz = 0. Now the three variables @, y, 2 are subject to one condition; their three differentials are therefore subject to one linear condition, so that two differentials may be given any arbitrarily assigned values. In particular, dy, dz may be assigned arbitrarily. Hence the coefficients of dy, dz in the equation last written must both be zero, scala mtA=0, nyta=0, ‘To summarize, the four equations aty+z=1, yetrA=0, wz+A=0, ay+A=0, determine the four unknowns #, y, 2, A. ‘Multiply the last three equations by «, y, z, and add. Then Bayz +Ale+y +2) = 0, or Bu+A=0, UNDETERMINED MULTIPLIERS 83 Multiply corresponding sides of these three equations; then aty%s? = 2703, or ut = 2708, Now w is not zero, under the conditions of the problem (x, y, 2 non-zero). Henoe ete We may extend this illustration to show how to settle whether such points are maxima or minima, It is easy to prove that there is only one such point, namely, e=y=z=} Consider, then, a near point z=ttp, y=tt+g z= t+", where p, q, r are small. Since x+y+z = 1, we have ptqtr=0, Now usaye = (+7) (849 (847) =drtiptgtr)+ Marti + pq) +per. Suppose that p, q, r are so small that the product pgr may be neglected. Then wat p rater? Me +e+s)} =P te +r). Hence w has its locally greatest value at (J, }, §), and the value gy is therefore a maximum. We now apply these ideas more generally. To locate the maxima or minima of the function Wf ly ty 5% q) Of the n variables 4,2, ...,%q subject to the m relations (mwhere 2.91% are constants, prove n—m differentials may be given arbitrarily assigned values, In that z has a stationary value when p, z, y, q are in geometrical particular, dain; 42%n4.99 +++» da, may be assigned arbitrarily. Hence progression, the coefficients of diz,,,.,...,d2, in the equation last written must : isan aoaas ae cane pe oe 3, The variable z is determined as a function of 2, y by the ea ei ere eas ae Byte trys +3 = 0. Bn" nia One em ‘s Wind the values of £3, 53) arg, When 2 =y=2= 1, and prove Dear. tn gee = 5 that these values of z, y, z make the function n In all, then, we have the m+n equations B+ ley a minimum, P(E, Zp ---r%q) =O (6 =1,...,m), Of 4 thy ebm ae 4, Find the minimum value of aj ae, tat Amey =0 G=1,....m) a+y*+(act+by+e)*.86 MAXIMA AND MINIMA REVISION EXAMPLES XI 87 5. Find all the stationary values of the function 11. The sum of the twelve edges of a rectangular block is &; the sum of the areas of the six faces is ,jh*. Prove that, when the excess = y_(e-y of the volume of the block over that of a cube, whose edge is equal vee a to the least edge of the block, is greatest, the least edge is zsh; of the variables , y; discuss how their nature depends on the value and find the other edges. of the parameter a, Consider in particular the points at which 13; “Find the grsatt isd Last abtenoos from the origin of ete: point on the surface 6. Find the greatest volume of a rectangular parallelepiped @ +()’ +()’- 1 which can be placed inside an ellipsoid of semi-axes a, 6, ¢ with its a) Nb ’ edges parallel to the axes. where a, b, c are fixed positive numbers and p is a fixed even integer [The edges 2x, 2y, 22 of the box are subject to the condition greater than 2, 2A #1, [The square of the distance is 2*+y*+2*.] IER iD as 13, Prove that, if a, f, y are positive, there exists a shape of 7. Find the maximum and minimum values, for real values of triangle for which the maximum value of 2, y, 2, of the quantity 2*+y*+2* subject to the conditions that sint A sin’ B sin’ 0 la+my+nz = 0, is attained; and that for this triangle ax? + by? +022 = 1, tant = 242+) tansy MOtP+Y) tone = WEtB+N where a, b, ¢ are positive and J, m, n are real. fy ae, ei Verify that the values determined are real and positive. 14, Find the stationary values of 8, Find the maximum and minimum values of the function y+ 422 — dyz—2en— Day uaattyte, subject to 2c? + 8y2+ 62? = 1. where z, y, z are connected by the relations 15. Show that, if x, y, z are connected by the relations at+ytz=a, @+y4+2=1, w+ye+22 = a le-+my+nz = 0, 9. If az+by-+cz=1 (a,b,¢ positive) and a, 6, care not all equal, then the extreme values of the function show that the values of 2, y, z for which V sax? + by? + cz* aes are the roots of the equation atytz r is stationary are given by 7 aa? = by? = o28, Show that this is a true maximum or minimum if zyz> 0, 16. On the surface given by a+y2tot— 2x4 2y+62+9 = 0, 10. Use Lagrange’s method of undetermined multipliers to a fi BOE SL Lo; show that the triangle of maximum area which can be inscribed in id the stationary points of the function a*+y?+2*—yz—z2z—2y, a given circle is equilateral. and investigate their nature,