Академический Документы
Профессиональный Документы
Культура Документы
arb’, ayn, % arb,
va re ma a
vanishing simultaneously in the range. Therefore the range must not extend 80
far as the first value of x, greater than the initial value 2, for which these
“four quantities simultaneously vanish. Thus we have a possible upper limit
for the range, and denote it by 2, when there exists such a value of «; and
then the Jacobi test makes ©, the upper limit of a range which has 2, for its
lower limit. ‘The range of the integral, consequently, is not to extend so far
as the conjugate of the lower limit.
We must obtain the analytical determination of the upper limit «,, to be
taken as the conjugate of z, the lower limit.
Critical equation, in connection with the Jacobi test.
202. The four integral-sets, 7, and {, 7 and & m and &, mand £4, are
linearly expressible (§ 184) in terms of the eight fundamental intogral-sets
br(2) and y, (2), for r=1, ..., 8. Let
n= 2 bb b= 3 bevel)
2h6@ b- Zt ee
so Emo
a = 1, r(2),
‘The quantities ey and xf, in each instance, denote a small variation from the
characteristic curve under consideration, chosen so as to give a contiguous
characteristic curve, For every curve, the constants are determined by initial
values chosen at the lower limit. Accordingly, we choose
8
= m Yr (2),
=, &=0; m=0, Sm
for r =1, 2, 3, 4, at the lower limit z,—a selection which makes each of the
four contiguous curves touch the fundamental characteristic curve at that
lower limit, Among these four contiguous curves, we discriminate by the
assignment of other initial conditions at that place # =<), as follows:
Gi" =0, a" =0)
hi" =o, a" =0 :
g"=0, cee
w= op308 CRITICAL FUNCTION IN (cx. vi
where p, 0, 7, # ate finite non-zero constants, Manifestly, the curves are
distinet from one another; and each of them is distinet from the character-
istic eurve which (§ 182) would arise, if either p or ¢ or + or # were to vanish.
All these assigned conditions are in accord with the six relations of § 188,
satisfied by the four integral-sets », and £, (for r=1, 2,8, 4). ‘Those relations
require that certain constants are to be zero: when the foregoing values are
substituted, they do, in fact, make each of the said constants equal to zero.
‘The eight constants hj, ..., ks in the expressions for n, and f, are thus
subject to the eight linear relations
Lhrbr (%)=0, Sher (@)=0, TkrG (@)=0, Lkeyr’ (4) =0,
hye” (a) = py Ser" (as)= 0, Serge" (a) = 0, Shep” (m) = 0.
The determinant of the coefficients of the constants & on the left-hand sides
1b (oo), Yoo), os’ (xa), Pu (ods Ba (to), Yo" (ads i” (ae), Yo" (a) |
is the value, at ay, of the quantity V of § 186, a quantity there proved not to
vanish anywhere in the range: we denote this value by W.. Let [£] denote
the minor of any constituent & in the determinantal expression for Vs, s0
that, for example,
[Gx” Co) = |e e), Yu wo), ui (eds Woe! (wads Ye" (we), 1” (eo), Yu” (He) |
[Gs (ood) =| Gu (0), Yow), be (ad, Yu! ods Ya" (aa)s fr” (eds Ws” (ao) |»
and so on. ‘Then we have
kWo= p[ di" (a)], ba ¥o
and so on: generally
Lda" (aa) Ka o= p [spa" (ae)], ba¥o= ps" (as)),
ke Mo= ploy” (a),
., 8. Similarly, for the same values of r in succession,
— LWo=o[hr" (a),
mVy= 1 [$e (wa),
— me Vo= ely" (a)
We therefore have unique finite non-zero values for all the coefficients in the
four integral-sets 7 and f, m and &, m and %, and m and £..
for r=1,
203. Now consider, once more, the quantity @, where
@=|%, tm m %
Gs Sar bs
me Ws ms me
oo Bw
Tt vanishes at the lower limit 2, of the integral; but no difficulty is thereby
caused in the second variation, because
La Y pa 40 +7V +82, Sak" + OV 492 +yV + xh208) ‘THE JACOBI TEST 309
It must not again vanish within the range of integration; and that range
must extend 90 far as the first value «, of e which is greater than 2, at which
the quantity could vanish. But
(p07) Y=
Zhe @)[e" (wd), Ede LY" (], Ede (a) [be CH), Shr (Ye Cod] |
Eabr (2) Ube" CW) Se Le" (wal, rr (@) Lbs” Ceo], Er EY” (#o)]
Ege’ (©) Ee" (oY Zhe! (w) Le” (ae)], Ede’ (a) (Ge (a) Zhe! (@) Le” (@o)]
| Eee [Ge ed], Se’ Ye" (aod), Ed’ Ube” od], Ee) Le”
‘The determinant on the right-hand side can be evaluated by the repeated
use of the known theorem relating to determinants made up of constituents
which are the complementary minors of determinants, themselves made up
of the constituents of a determinant such as Vo. Thus, in the expansion of
the foregoing determinantal expression for @, the coefficient of
dH @), br) $s (@), br (2)
WE), HO), Ww, We)
Gi (2), by (@), di(e), di (@)
WO IO WO, We
[d."” GL [oe” Cd, [4s Gd] [he Co] | -
Gh" @) Be Gl, Ge" @L, WH @))
[i (em) C6" od], [ps [40 @))
IC) a CC AC) RC
By the theorem quoted, this last determinant (made up of first minors of Y.)
is equal to
We] be (a), he (@), gr (a), he (We) |
Vs (2), Yo), Vr (me), Ye (aa)
$e (0), ge (i), (ao), s' (aed |
We (ea), We (od, Yee), es (a) |
Similarly for other terms in the expansion : the coefficient of
Ide(2), vale), oF (@), Wi ()\,
where p, g, r, &, are any combination from 1, 2, 3, 4, 5, 6, 7, 8, is
We | peal), Ver) Fra(m) Yor (m)|,
by the same theorem. Consequently, when all the evaluated terms are added
together, we have
Hime, Yale), G5), WE), bo (ee), Pole), 1’ (Be) Ys (2)|
=A(@, 2),310 EXCLUSIVE CONJUGATE-BOUNDED RANGES [on v1
where A(#, 2) denotes the determinant of eight rows, made up of $(2),
¥(@), $(@), («,), and their first derivatives.
Now ®, does not vanich, nor become infinite, The quantities p, ¢, r, p are
finite quantities, respectively determining the four characteristics contiguous
to the characteristic under consideration. The quantity is not to vanish
along the range; henee the function A(e,, 2) must not vanish along the
range for any value of # within the range greater than a.
The conjugate of a, is given by that value of «, say , for which
a i
Zam=0, Sato, Zan'=0, 3 at'=0,
without a, dy, ds, 4 themsolves vanishing: that is, for which @ vanishes.
Hence the conjugate, given by «,, of the lower limit of the range of integration
given by ay, is the first root (greater than 2,) of the equation
Ale, a) =0.
If the original integral is to possess a maximum or a minimum, the range of
that integral, beginning at #,, must not extend as far as the value =, which
defines the conjugate of ~ on the characteristic curve.
We thus have the third test for the possession of a maximum or minimum;
afver the analogy of preceding instances, it is called the Jacobi test.
A range, bounded by two conjugates, does not enclose any similarly
bounded range.
204. In enunciating this result, one property has been tacitly assumed by
anticipation: viz. a complete range along a characteristic curve, between
point and its conjugate, cannot contain within itself another complete range
of the same type: in other words, the conjugate of any point within a range
lies without the range, The establishment of this property can be effected as
in the preceding cases (&§ 119, 176), by proceeding from the function A (2, 2),
the outline being as follows,
This function A(e%, 2) does not vanish anywhere that lies actually
between «, and ite conjugate «,. Let two points be taken in the range, one
near #, given by a+£, (where X and £, are positive, and X is small), and one
near «, given by 2, ~ e£, (where ¢ and f, are positive, and ¢ is small); the two
magnitudes
Ay, e+ dfs), A (a, a — ef)
have the same sign. But, approximately, for sufficiently small values of 2,
we have
A (oa, tot Mh) = EME Wo,
while
Ae, 8h) =A (es, a) ef 8 SH)206) SUMMARY OF TESTS ail
so that
©) has a persistent sign, opposite to the persistent sign of ¥
a
whatever initial point be chosen. Also, as A(#, #,—ef,) does not vanish,
a Gu8) does not vanish.
When a point a, +a£ is taken, with a and & positive, and ite conjugate
4, +a€' is wanted, a being small, we have
A(e+ab, 2, +08’) = 0;
so that, when a is sufficiently small, we have
20. (a, 21) pO (ae,
eat
If £ and £’ have opposite signs, the new range would lie within the old; and
the distances of the new ends from the old ends would be of the same order of
magnitude. ‘This diminished range would similarly lead to a more restricted
range, because of the persistence of the sign of V. Gradually, as before, we
could wear down the range (on the hypothesis that £ and £’ had opposite
signs) until the fanetion A(z’, 2”) would palpably cease to vanish when 2” is
sufficiently near a’. ‘That hypothesis must therefore be abandoned. Con-
sequently, & and £ have the same sign; and they are of the same order of
magnitude.
Proceeding gradually from 2,, and taking the new conjugate for each new
initial point—which conjugate must lie beyond each immediately earlier con-
jugate—wo infor the property that the conjugate of any point within a
complete range lies outside the range.
=0.
Summary of tests.
205. The conditions, when satisfied, secure the existence of a maximum
or a minimum for weak variations. The Euler test and the terminal con-
ditions make the first variation vanish. The Legendre test and the Jacobi
test preserve one sign for the second variation—either positive (when there
is a minimum) or negative (when there is a maximum)—through all weak
variations that are possible.
General weak variations : statement of results concerning the ‘second’ variation.
206. When general weak variations are considered, so that we have to
deal with the integral
[Pee 215 Ye Yes se DME,
and the variations are «+ «X, y+#Y, 2-+Z, the results connected with the
first variation of that integral have already been obtained. Much laborious
algebra is required for the reduction of its second variation to a normal form.312 SYNOPSIS OF RESULTS FOR [en. vi
The results are significant, not least in connection with the discussion of a
characteristic curve, of characteristics contiguous to that curve, and of a range
along the curve bounded by conjugate points, with the definition of conjugates
already used. The discussion follows, in more general shape, the discussion
already given ; and the algebraical calculations, in themselves and in their
process, are the extension—an elaborate extension—of calculations of the
same kind already effected for simpler cases. It may, therefore, suffice to give
an outline only, arranged in successive statements, ‘Three fundamental
quantities are definitely established, in (II); the other inferences are stated,
without detailed proofs being given.
(D The function F satisfies two identities
2F yy FoF
PF. ak sy Ben homade Fs oy oe Pte ge
Derivatives of F with regard to the variables
2 By Bey Ys Yr Yoo Bs fav Bn
are denoted by numerical suffixes 1, 2, 8, 4,5, 6,7, 8, 9, respectively; and the
variable quantities
X, XX ¥,V, Va, Z 4 Zn
are denoted respectively by
8, Ox, Bs, Oey B55 Bo, Ory Bar Bs
‘We shall want combinations of Y and X, of Z and X, and the derivatives
of these combinations; we write
b=V =m¥-yX=28- 4,
bs Vi = 2,0, 91004 (00, —Y201)
= ,0,— Ya + 2 (08.— 204) + (0, ~ 8),
2 = 5,X=2,0,-2,8,,
8, — £18, + (048, — 2,84),
— 2,0, + 2 (0, — 2405) + (040, ~ 248,)-
(ID) Differentiating the first of the identities satisfied by F with respect
to a, to y, and to z,, separately, we have
Fat Fut 4Fa=0,
Fg + iF t Fy =0,
2,F at \Fat iFa=0.
Consequently, there exist three quantities gx, ga, gu such that
Fugu, Fo= gm, Po= 2m
Fy =—2trga-%%9e, Fe=- GYgu— 22 Go,
Fes= y'gut 2yr29u0t 2790
a206] (GENERAL WEAK VARIATIONS 318
‘The three quantities gy, ga» Jw are fundamental; they are, save as to a power
of z,, equal to the quantities fr, fy fis of & 182—5.
(AIL) The first variation of the integral has been made to vanish, thereby
leading to the chateeteristic equations and utilising the terminal data so that
the limits of the integral ean be considered as fixed.
‘The second variation of the integral is ie fo dt, taken between the fixed
limits, where
= EE FanOnOns
the full expression of © being Fy0°+ 2F 40,0, +...+ 2Fixh6.+ F,8% We
take a quantity
(C= AO, + BO: + 08; + 20,0, + 20,0, + 20,0,
+00 + 502+ 00; + 2/00, + 2980, + 20,0,
+ 20, (a0, + 0,-+ 70.) +28,(NO,+ 105+ v0,) + 26, (pb, +08, +76,).
‘The first aim of the analysis is to modify the second variation, by expressing
@+ a as equal to ®, where
and, in ®, as few terms are retained as possible. For then
le +90) a= fou +(a)= [ow
because the limits are now fixed and therefore every variable 8 in [] vanishes
at each limit (0,, 0,, 0, are absent from []): that is,
fou=foa.
‘Thus © can be regarded as an intermediate normal quadratic form for the
integral,
‘As regards the retention of as few terms in ® as possible, we have
m6 W+ 2g WW" +2941 W" +g + 296 W' 4 gy?
=G0W" + Gea Gu’ — 2m) W? + (Gn — Gs + gn’) W?
+H iggW+ 20m WW’ + (oun — gn’) Wh
The term in the last line, with its sign changed, can be moved into 2, and
then only the squared terms in the first line will be left: that is, by a change
of the coefficients gnn which initially are at our disposal, and without loss of
generality, we can choose
9o=% ga=0, gn=0.314 SYNOPSIS OF RESULTS FOR cu. vi
Similarly, without loss of generality, we can take
ga=0, ga=, ga=0,
in the form initially postulated for &.
We therefore can take, in umbral notation,
D= (he + rede + 11h + Mobs + Hobs + Yeh)?
‘The six umbral symbols 7%, 76, Ye» Y Ye Ye combine quadratically into the
significance of reality, according to the laws
Sous
(the three fundamental quantities already introduced) ;
Hr=0, HH=9, NH=0, YW N=, WE
(according to the foregoing initial simplifications of ); and, for all other
combinations,
Yn Yn = Irons
these non-zero coefficients gaa remaining.
The consequent modifications, by the absorption of
= [G0 + 2900 W’ + (Gr— go’) WH
into [, will be supposed made, so that the coefficients in [] (as yet unused)
will be regarded as having absorbed the necessary changes,
(IV) Owing to the expressions for the real variables ¢ in terms of the
real variables 0, we have
= Yor Bo + (Zroite + yet) 85 + (Yates + Yetta + 11%) By
3m
F 6,85 + (Quite + Yes) Os + (Ys + Yatta+ Yet) A
— (m2 + Yeh) Bs
— Qyots + Wey + W621 + Ys) Oe
= (nts t Ys + V620+ Toye t He +1) Or
‘Then we are to have ,
os ae-{ nel)
where, on the right-hand side, we substitute the equivalent expression which
is quadratic in 6,
(V) Equating the coefficients of the various combinations of the variables
64, --+, 8 on the two sides of the last relation in (IV), we have relations
between the known coofficients Fis, the unknown coofficients gman, and the
coefficients in C]. ‘The total number of relations in this aggregate is forty-five,
being the number of coefficients Fn,206) GENERAL WEAK VARIATIONS 315
(i) The terms, involving (6,, 64, 6) in @ +29 and &, agree without
jt
any relation, owing to the expressions for Fg, (with m, n=3,6, 9) in terms of
‘the three coeflicients gun, Jer Joe:
Gi) The terms, involving (6,, 0., 8,%0., 6, 64) agree, if
Fat C= %wo.t% }
Jays
FotF= %at.te+ Gut?
Pat =~ 2gnt22— 2getiyYs
Fat P= 2gux%+ gar
Fut B= gure, i
Fug + H= — get 24 2G Yo ~ Joa 5
Fat G = — 2gnti4— Bayi —Gatyy
— yen &s— Wears — Joss
gotta + 2a (sige t ther) + 2gariss |”
+ 90h t Inrith
In these relations, F occurs twice, G twice, H twice; and their repeated
values must be the same. But from the fundamental identities, we have
oF
Fat Fut Fam — Se
Fy + yiPrg + 2 Fu + 20, Pre + 2ysFun + 2e4Fs z.
which give
aA tyHtnGa ee,
and
I: Fas + Fos) + 2, Fo — Fs) = 2 (t4Fes + YsFrs + #2F x).
Similarly
Hay B+ Pa Se
and
41 (Fa~ Fa) + (Fru Fy) =2 (0 F gt oP + 4F,
and
G+ y P4200,
and
@, (Fy — Fa) + ys (Fes — Fea) = 2 (Fon + Ys Fen + 2 Fy).
Of these necessary equations, the three which are free from A, B, 0, F, @, H,
are satisfied by the foregoing nine postulated equations, if only
Fea— Fa= (Guu) 2?
‘Therefore the whole set of equations defines A, B, C, F, @, H, gus gu, always
subject to the latter equation,316 ‘TRANSFORMATION OF (on. vr
(iii) The terms, involving (6,, 8, 836,, 6, 6) agree, if
Fett = gutsy
Fotv = Gat %+ Got\het gat! | >
Fat =—Gutiks— Jo%Ys— Jaye Jas
Fate
Ja X; + Yeu, Xs + Jos?
Fat h= gute >
Fig t B= Grates ~ Gui Ys— Gute Jo%e
GooFs%s — Jeoyf%s— Joti Xe — JarTiYr
GeoFaks— Gstiits — oFikra— Ju
Goti2s+ Jo (Ys + is) +IaYiYs
+ Gato Jathls + Ioyit%i+ Jair
which imay be regarded as defining the nine quantities a, ..., 7, in terms of
new quantities gun, Jus Jer Jo
(iv) ‘The terms, involving (6, 6, 6)* agree, if
Fret OC +27 = Aguas + gua?
Pot PF ty to= Agee + putts + gate + gut?
Fat @' +4 +p
Agmitata— AgeeteYs
2x Yo— Wu Ys ite — Pours — Jest
Aguas + gut?
gees — 49a
— gusts &a— 2s Fe — Jour 2 — Jessa
FatA' 42a = 4gnz + Bgetitet Sguye |
+ Agents Ys + Agayits + Inti? + Wayizi + Ieys?
‘The quantities A, ..., H, a, ..., 7 have been defined in the earlier group (ii)
and the earlier group (iii) of relations. ‘These six conditions therefore define
$us Jus Joss and they leave three relations to be satisfied.
(v) The terms in (@,, 0, 8:16, 64, 6) agree, if
Fatt $6= 2gutets+ Qutits
Pat VAS = 2g e+ 2st + 2Gat is + Gusts + Joab + Gace
Fist 1 +9 = —2gmtty2s— 2guitays— 2gut — 2a Gite
= Ieatrea~ JesTrYs— IsTrYo— Jats
Fat ot f= Yutatts + 29+ 2atits + 2a 2+ Getta Int?
Fete +b = gutter t Jost, |
Fist B+ h= — 2get.25— 2geaitays — 2Gertata — gents | :
—Getsts— Guat te~ Goi Ys — Gorter
Fut B +p
Fat H+ B+206] SECOND VARIATION 317
Fat 0! +9 =—2gmtats— 2a yets — Ways — Waa
state Foss — Gui %s— Gertiyy
Fy th +h — 29a 2sits — 2grotrts — 2a te 2WostsYs
— Gusts Junta Gast Juha
Fatal +a= 2gntrts + 2ga (tryst Y%s) + 2geateYs
+ Gea 2yata+ Wits) + Gua (2yate + 414s)
+ GentiFet Jos (Ye + Yi22) + Gade
+ 2gathte + WatiYet+ Get In) Y%
(vi) The terms in (6,, 6, 6,)* agree if
Fate = goot? + gat! + guts
Fit f= gate + (gout Jor) Cats + (Gea + Ga) trees + Pout + (Jaa Gx) tra + Gat?
Frat J =~ Gnitsts— Jo%sYs— GnYe%s— Joye
— GIeatsta—Jesta— Jetafe— Jayate
= Gni%s81— Joris — Jato Iaith
Fy t B= gets + Gott + guts
Fu th! =—gaitsts— Gu%sYs— Jeaa%s— Jai %
= Geos£s— Jua¥s— Jer? Bs Jutate
= Gutits— Gu2a— Gosh ~ Jur
ta = Gute +2GaYs2st Poor + Gust + 2GrsZYot Ia + Inzt + Wait tIay?
F 2geeYats + Aue + WosyYs2s + Waza + Woyszr + Weyer
(vii) Various relations are satisfied by the quantities Fn, which arise as
derivatives of the two identities satisfied by the subject of integration, ‘Thus,
from the first of these identities, we have
Fy + Fat 4iFn
Oy Py + y Fut AF
2 Fat yFat Fs
and thence
204 yB+27=0,
mrt Wet AY =O,
aptyotar=0.
Again, from that same identity, we have
iF
an Fat late han- ab — uF —20,
=a Pgt Pot Fe
aH—yB-aF,
ena Fat Fat Fann d— nH —aG;
2G +P +50 +aG+yF +20
=F yt eka t %Pot Fat oka t Pot Fn t Fat km,
with two similar relations.318 SYNOPSIS OF RESULTS (on. vi
From the second identity, we have
0 = 0, Fat ¥: Fest 2 Fg + 20, F es + 2yeF eu + 2eF re,
Oma Fn t Yi Fot Feet WeF s+ 2yeFugt 26.F oy,
Oma Fn t Fat 2: Feat laa Fn + 24 Fest 25,Feo.
When the values of Fy, ..., Fin, Faso» Fare substituted here, these relations
are satisfied identically,
Similarly with the other relations, that can be derived from the two
identities in the same manner as those in Chapter II and Chapter III: it
can be verified that each of them is identically satisfied when the values
of Fyn are substituted,
(VI) There are subsidiary characteristic equations, associated with the
fandamental characteristic equations. Let 20 denote the same function of
BEL En 0,60, 0", a8 @ is of @,, ..., 0, with the coefficients Fyn. The
subsidiary equations axe
a _ d (an
ae ~ aloe’) * de
aaa) * 2¢f)-.
and em ad (8)
ea oe) +aelie)=2
which are really equivalent to two equations in virtue of the relations among
the coefficients.
(VID) Corresponding to V and W, we introduce the combinations
yan nk,
fab
We denote by 2T, the same function of y, yi, Ys, 2, %1, %, a8 ® is of
v,V’,¥", W, W, W". Then y and z satisfy the equations
rid @& at
oy ~ (op) * ae Gye) =®
ar det), & er
de ~ at (ie) +a (fer) =
which are the modified subsidiary equations.
‘The primitive of the subsidiary equations, and therefore also the primitive
of the modified subsidiary equations, can be derived directly from that of the
fandamental characteristic equations. Every integral-set is made up of linear
combinations, with constant coefficients, of eight linearly independent integral-
sets which constitute a fundamental system.
‘There are at least five such integral-sets, any four of which (in pair-
combinations) provide six relations, identical in form with the six relations of
§ 188: a selected four are denoted by y, and 2,, yz and z,, y,and z, y, and &,.
ao206) LEGENDRE TEST: JACOBI TEST 319
(VIE) By means of the relation
Iv"
fou= (e+) a—(v1
= fae,
where V is a homogeneous quadratic expression in V, V’, W, W’, we can (by
analysis exactly similar to the analysis in § 200) obtain an expression
(g0W* + 29. W¥ + Ju"),
where
wi 8 Yn By
Ye, Bs Yay Be
Ws Bs Yan Be
BL Ya
ee w=| ee
Yo 3 WB, Yh
Yoo te Wer By Yar Be
Yo", ss Bs Yar Be Yo, Bi, Yu My
VW. Ye, Ry Yu We Rs Yo Be
where @ is a quantity that does not vanish within the range of integration.
This form {tat is the final normal form, as {dt was the fist or inter-
mediate normal form, of the second variation.
(IX) The discussion of this result proceeds in the same way as the dis-
cussion in § 200—204,
The Legendre test requires that, everywhere in the range,
Guu > Ga"
When this is satisfied, a common positive sign for gu and gy admits a
minimum for the original integral, and a common negative sign for gu and gy
admits a maximum.
The Jacobi test requires that the range of integration, which begins at an
initial place t on the characteristic curve, shall not extend as far as the place
4, the conjugate of t: where @; is the first value of (greater than t)at which
the determinant
[yi (G), 2'(f), Yolts Bt) Yel (to), Be (tds ¥e(4), a(t) |
vanishes.
(X) A complete range on the characteristic curve, bounded by a place
and its conjugate, does ot contain within itself another complete range.CHAPTER VIL
ORDINARY INTEGRALS UNDER STRONG VARIATIONS, AND THE
Wererstrass Test: Soup OF LEAST RESISTANCE: ACTION.
General notion of a small variation.
207. The systematic analysis dealing with irregular small variations,
which can be represented by a jagged line resoluble into small straight
pieces, is due to Weierstrass, He introduced it (and developed it, thereby
meeting some general objections of Steiner) so as to discuss possible maxima
and minima of integrals, which involve one origioal dependent variable
and its first derivative. Jagged lines were recognised by Legendre* as
possibly leading—in the case of Newton’s problem of the solid resistance, he
obtained jagged lines which do lead—to results, reduced below the minimum}
provided by the characteristic curve under small variations that now are
called ‘weak.’ Such lines of a zig-zag type were excluded from consideration
in their analysis by Legendre and succeeding generations of writers, through
@ definition which was implicit at first and was made explicit only after-
wards. It had been assumed, usually without any statement or only on a
later definite statement kept unobtrusive, that, when the variation of the
dependent variable y is maintained small, the variation of its derivative y/
will also be small. The assumption seemed to be that the smaliness of y’,
consequent upon the smallness of y, was too obvious to merit even mention.
‘There was the further assumption, also implicit and ignored in statement,
that the variation of y’ is to be of the same small order as the small variation
of y. All such variations were represented by by, y, Sdy, and so on; the
mere presence of the Lagrange symbol 8, in this association, apparently
ensured the requirement of smallness.
Moreover, in spite of restrictions in the range of the integral to small
variations dy of the ordinate alone, circumstances compelled the consideration
of small variations of « at the limits of the range (as at mobile limits, in
§ 81), though these were not admitted elsewhere ; and a compromise had to
be framed, expedient for the discussion of such limits,
* “Mémoire sur la maniére de distinguer lea maxima des minima dans le caleal de variations,”
Hist, de V Acad. Roy. des Sciences (1788),
‘f Legendre (1c.) called # maximom or a minit208] GENERAL SMALL VARIATION 321
Manifestly the general notion of the slight deformation of a curve, into a
contiguous curve, merely requires in essence that the distance, between the
position of a point on the curve and the displaced position of the corre-
sponding point on the deformed curve, must always be small, whatever point
be chosen, If this requirement be met, the contiguous curve gives a small
variation of the original curve, whatever be its shape in detail. We are
accustomed, over a stretch of continuous curve, to very rapid small changes
which, while of full importance in detail, would be deemed of slight statistical
account as compared with the smoothness of an averaging curve over that
stretch. Restrictions at every point, as to changes of direction, of curvature
and the like, that all of these changes shall be small when the variation is
small, are essential limitations upon the completeness of the variations
admitted or imposed. A minimum for restricted (weak) small variations,
which fails to be a minimum for unrestricted (strong) small variations, cannot
be claimed as a true minimum, any more than the summit of a mountain
pass could be regarded as a true minimum, because there is no higher point
to which a pedestrian need rise when passing from the valley on one side
of the pass to the valley on the other side, or as a true maximum, becanse
there is no lower point to which a pedestrian need descend when passing
from the ridge on one side of the pass to the ridge on the other side. A true
minimum and a true maximum must possess their distinctive properties, in
all circumstances of variation, and not merely for a selected group of such
circumstances,
Accordingly, we proceed to the consideration of strong variations, indi-
cating by the term specially the types that can be resolved into an aggregate
of variations of a specifically simple character. The character of these com-
ponent simple strong variations must now be explained.
Strong variations : fundamental constituent type.
208. In all the preceding investigations, the small variations of the
characteristic curve have been of the type styled weak. ‘They are such that,
if the point © +f, y +47 be the varied position of a point @, y on a curve,
for which « and y are expressed as functions of a continuously increasing
parameter ¢, the quantities £ and are arbitrary functions of ¢, independent
of and limited by the requirement of being everywhere regular functions,
so that they possess their due succession of derivatives which also are regular
functions,
We now proceed to consider small variations which are not limited to
this class. But the mere exclusion of such limitation would be mainly a
negative description. For the purposes of mathematical representation, it is
convenient to select particular types of small variation from among those
which hitherto have been excluded from consideration. Thus we have
already (§ 47, 102) had small variations, where isolated discontinuities of322 NON-REGULAR SMALL VARIATIONS (ow. vir
direction within a range have been admissible; though the number of such
discontinuities in any finite range has been limited, owing to the require-
ment of isolation. We have mentioned, though only by the way of illus-
tration of possible small variations that are not of the definite weak type,
the instance
= wasin(te"),
Such a quantity & is of course finite, continuous, and small when « is small
(vis a constant), But the first derivative of & is of an order smaller than
the order of £ when n is positive. When n is negative, the derivative of £ is
of an order not so small as that of £; it may be finite, or it may even be
Jarge and oscillate rapidly, when « is small. In all such instances, the
expansion of the subject of integration in powers of «, after the small
variation has been imposed, would not be effective to the end for which it
was to be used. Again, it is conceivable that we could have small variations
over a range continuous, not merely in are but also in direction, yet not
continuous in curvature; an integrand of the second order would suffer a dis-
continuity in passing through such a place, even if the place were isolated ;
and difficulties would arise over the curvature properties of the contiguous
varied curve. Similarly, if the subject of integration involves derivatives of
order higher than the second, we should be faced with the possibility of
abrupt changes in the rate of curvature.
209. An indication of one method, of taking some of such non-regular
small variations into account, is given by the consideration of an integral
2. ae
‘
with a jagged variation, Let AMPRS... be a characteristic curve, on which
A’ (the conjugate of A) lies beyond 8. Let any point P in this range AS
be displaced to a neighbouring place Q. Because P lies within the range
bounded by A and its conjugate A’, a consecutive characteristic AQT...
can be drawn (§ 70) from A to pass through Q; and this consecutive curve
does not meet AMPRS... until A’.
Suppose that the conditions for a minimum are required. (The alter-
native conditions for a maximum follow by changing, throughout the ex-
planation, a greater inequality into a lesser inequality.) Accordingly, if the
integral is to be a minimum along a characteristic curve AMP from A to P,
and if we denote its value by J,» when taken along that curve, we must have
Luxe Toe > Lanes210) ‘REQUIREMENT FOR CONSTITUENT JAGGED VARIATION 323,
where
(Lexa is the integral along ANQ from A to Q;
(ii) Zgp is the integral along QP, the value of the independent variable
of integration along QP increasing from Q up to P; and
(iii) Typ is the integral along AMP from A to P.
Now ANQT... equally is a characteristic curve, determined by slightly
different initial conditions. Equally it provides a minimum: so that, taking
AMPQ as a variation of ANQ, we have
Laue + Inq > Lana
where the integrals I4yp and I4yq are the same as before, and Iyq is the
integral along PQ, the value of the independent variable of integration
along PQ increasing from P up to Q (The integral Zpg is not the integral
Top reversed: along PQ in Ipg, the independent variable of integration is
increasing and likewise, along QP in Igp, it is increasing.)
Next, consider ANQ and QR as a variation of the continuous length
AMPR. Owing to the requirement of a minimum, we have
Laxa+ Ton > Laure,
where Igyq is the same as before, Igy is the integral from Q to R along QR,
and I,ypx is the integral along the characteristic are AMPR. Thus
Laur + Inq + Ton > Lava + Ton > Laser:
On the left-hand side, the sum of the integrals is the whole integral, first
along AMP from A to P, next along PQ from P to Q, and then along QR
from @ to R. Thus the integral along AMPQR is greater than the integral
along AMPR, Hence if the minimum condition is obeyed for a broken line
AMPQ as a variation of a characteristic are AQ, and for a broken line
ANQR as a variation of a characteristic are AR, it is obeyed for a line
AMPQR, with a jagged portion PQR, as a variation of the characteristic
are AMPR.
It follows that the minimum condition will be secured for any jagged
variation of a characteristic arc, composed of portions such as PQR which may
be taken on either side of the are AMPR..., when a requirement is satisfied,
that the minimum condition holds along any characteristic curve, separately
for each broken curve, suck as ANQP for AMP, and such as ANQR for
AMPR. In other words, the requirement is to be satisfied, whatever point P
bbe chosen on a characteristic arc, and whatever small are be drawn from that
chosen point to any contiguous point Q.
210. It should be noted that considerable freedom, thus far, is allowed
to the form of the small ares, such as PQ and QR. No assumption has been
made, tacitly or overtly, that they are straight lines: but we have assumed324 CONSTITUENT TYPE OF STRONG VARIATION [ow. vir
(tacitly) that such an are PQ is continuous in arc-length, and that it does
not eut the characteristic curve except at P. The assumption of continuity
of are-length in the variation remains an assumption. ‘The other assumption
is not essential: for, if PQ erossed the curve at points U, V,..., we should
consider PQ resolved into portions PU, UV,..., each similar to PQR in the
figure; and then we should treat each portion in the same way as PQR.
We do not therefore need to modify the requirement of the minimum con-
dition for a broken line such as AQP for AP, where P is any point on the
line and PQ is any direction through P.
Further, there need be no assumption as to continuity of direction, or of
curvature, and so on, in the broken or jagged variation. ‘Thus consider
acurve such as PQSTR. We can
draw characteristics AQ, AS, AT,
consecutive to APR. When the
minimum condition holds for a
single broken line such as AQP
taken asa variation of AP, for every
point on any characteristic curve and for every direction through the point,
it holds generally. For, with the preceding notation, we have
Laue + Tra > Lanes
Taxa + Tos > Lass
Lys +1 en> Lar
Lap +Iyn> Lascen’
and therefore
Laan Ira Los + Tor + Ten > Lowen
Hence the minimum condition will hold for any accumulation of broken
jagged small variations such as PQSTR, as it will hold for even the simplest
variation such as the broken line AQP for AP, if the minimum condition
be obeyed for each such simplest variation from every point P on the general
characteristic curve to any contiguous point Q off the are.
Note. As already stated, all the explanations and all the results hold as
regards the conditions requisite for a possible maximum, provided the greater
inequalities are changed into lesser inequalities.
‘Strong variation of ordinary integral with first derivatives.
211. Variations of the nature indicated are called strong variations. It is
to be noted that they are only one type of the variations, distinct in their
character from those which have hitherto been considered. The explanation
just given shews that some of these strong variations can be compounded of
‘variations of the simple broken-line type, while each individual component of212) STRONG VARIATION OF INTEGRAL OF FIRST ORDER 325
the composite variation is itself a possible variant of the central curve.
Further, when the condition for a minimum (or a maximum) holds at every
place on that curve and for every direction at that place, the effect of the
variation becomes cumulative. ‘The condition for a minimum (ora maximum)
is satisfied for a quite general composite strong variation, if it is always
satisfied for the quite general ultimate constituent of fundamental type.
We therefore proceed to consider the analytical expression of this con-
dition, initially individual for @ broken Jine, and scoure in its cumulative
effect. As the introduction of strong variations, and the expre
condition for the simplest and most frequent type of such variations, are due
to Weierstrass, the analytical requirement for the fundamental constituent
strong variation may fitly be called the Weierstrass test, It ean be obtained
as follows for the simplest kind of integral.
212. Let a point Q, with coordinates «+ x£, y-+1, be the displaced
Position of a point P on the central characteristic curve. The are PQ joining
Pand Q is not restricted, save that it shall be continuous in the intrinsic
qualities (such as tangency, curvature, rate of curvature, and so on) repre-
sented by the derivatives of the dependent variable which occur in the
subject of integration. If, and when, there should come a cessation of eon-
tinuity in any of those properties, the are in question would be held to cease
at such a point 7’; a new consecutive charactoristie A’ would be drawn
through 7, and a new are would be considered to begin at 7. Thus within
the limited range PQ the quantities £ and » will be regarded as regular
fanctions of some parameter along that arc, valid only up to @. For variations
beyond Q, new functions £ and », similarly regular along a new limited are,
will be required, and so on: the test will be demanded in connection with
each such elementary are.
It is to bo noted that all such variations are of a restricted type; thus
they do not cover variations such as
xcos(nte~4), resin (nte”}),
to take simple functions in terms of which, by Fourier series, some less simple
variations can be represented.
We shall deal, in the first instance, with the simplest kind of integral
‘
T= fF yaya
already considered in Chapter II; and we shall assume that the same con-
ditions, as before, attach to the integral. Also,all the tests, there found sufficient
and necessary to secure a maximum or a minimum for weak variations, will be
supposed to be already satisfied. We now have to consider the combination
Taxa + Top — Tames326 ‘STRONG VARIATION OF [en. vir
where AMP is the central characteristic; ANQ is the consecutive charac-
teristic through Q a point contiguous to P, provided (§ 70—72) P is a point
within a range AA’ bounded by the initial point A and its conjugate A’
g
—
x
By definition, Zgp is the integral taken from Q to P along any curve joining
Pand Q, continuous in are, direction, curvature and the like properties, but
not meeting AMP save at P.
Denoting the small weak variation from any point Mf on AMP to the cor-
responding point Non AQ by xu and ev, and regarding the integral Luo
as the variation of I4yp, we have (§ 41, 45),
or ayy
?
te [Few- smears [« (Ree +l
where [1] denotes the aggregate of terms of the second and higher powers
of small quantities. Now u=0, v=0,at A; and u=£,v=m, at P: also
everywhere along the characteristic curve, the characteristic equation @ = 0
is satisfied. Hence, neglecting terms of the second and higher orders as com-
pared with terms of the first order, we have
cr)
Tayo ~ Lane = « (ERE
213. The function F(z, y, 2, y:), in its initial form which involves
2, y, %, y, a8 unknown functions of f, is not a perfect differential with respect
tot; and therefore the integral Igp, being
[Pexewe,
7
cannot depend solely upon the values at P and at Q, but will depend partly
upon the path from Q to P.
‘The simplest curve to select is the straight line joining Q to P. We shall
therefore begin with the case when the are QP is a straight line. Let 4 and
be the direction-cosines of the line QP in the direction from Q to P; and
let U denote the length of PQ, so that
eE=—M, eg =— pl
‘Also, let be any point on QP, and let QS=o, so that ¢ may range from 0
to Lalong the path QP from Q to P. The coordinates of 8 are
a-dl-e) y-w(l-e).213) INTEGRAL OF FIRST ORDER 327
As the integral |F(e, y, %, :)d¢ is invariantive (§ 38), so far as concerns
gral nH
change of the independent variable ¢ of integration, we shall take «as the
current independent variable of integration for Jp. AS NOW 2,=2, y,=1,
along QP, we have
Zap |Fle-N(t=0), y~n(l—o), ® ph de,
while 1 is a small quantity of the same order of magnitude as x. The function
F is regular over the range of integration which, moreover, is small. Hence,
by the first theorem of mean value, there is a positive proper fraction ¢ such
that
Top = UF (a— nel, y ~ pel, , u)
=IF (ay, % w)+[P),
where [J*] denotes an aggregate of quantities of the second and higher orders
in |, that is, in , Consequently, up to small quantities of the first order, we
can take
Top =1F (2, ys >, p).
‘Thus
OF (@,
Lara + Tor ~ Tasen= 1 {P(g uw) —92F
accurate up to the first order. Now
aF (ey, a my
oy
FE ya yaa FEMA yy Mess
identically ; and therefore
F@ydm=r hyde)
Hace
Tavat Tor Lane = UE (e, y, 5 Yrs wy
where
Ey Yu w= Fe y, dw) —V OE
P(e. yu) _IF(e, y, ew)
=a ee +f
This quantity E(x, y, #., 1,2, w) depends, partly upon the position «, y
on the characteristic curve and the direction «,, y, of the tangent to the
characteristic curve at that position, and partly upon the direction 2, 4 of
the short line PQ through that position: so that, if there is a variation off
the curve, PQ must not coincide in direction with the tangent at P. Further,
F(a, y, %1, yy) is homogeneous of the first degree in 2, and y,; and therefore
PY YW) ang OF Ys Hs th
en, ay,
order zero. ‘They are unchanged in value when «, and y, are multiplied
are homogeneous in those quantities of328 WEIERSTRASS TEST fox. vin
by the same quantity; hence, when we multiply them by : so that they
vecune
istic eurve, the derivatives become 22, P(e, yay’) and é F(ey,2,y). We
now have
EO 2 JW =F BAZ PH 8 HP H HY)
ae ay
(=a) and w (=y), where s is the are-length along the character-
of reyan renin te remw-g reser}
The Weierstrass E-function, and Weierstrass test.
214. The integral J is to be a maximum or minimum for all small
variations. The conditions for weak variations have been obtained and are
satisfied. If the integral is to be a minimum under the strong variations
composed of variations such as ANQP for AMP, then
E(@, Yo YB)
must be positive, for all positions «, y on the characteristic curve, and for
all directions 2, through every position giving a variation, that is, for all
directions off the tangent ; while, if the integral is to be a maximum, the
same function must be negative for all positions 2, y on the characteristic
curve, and for all non-tangential directions through every position.
‘The function 14 measures the excess over @ minimum or the deficiency
from a maximum. Conversely, if Eis always positive, an excess always arises,
so that there isa minimum ; while, if Z is always negative, a deficiency always
arises, so that there is a maximum. ‘The function is often called the excess-
function, though it is a deficiency when a maximum exists. As it is due to
Weierstrass, it will be called the Weierstrass E-function. The test—that it is
to be always positive for a minimum, and always negative for a maximum,
along the characteristic curve—will be called the Weierstrass test. Manifestly,
it is distinct in character, as in source, from the Buler test, the Legendre test,
and the Jacobi test. The requirement is that the quantity
US Ferm wrens nh
i
+H 1S Few mn BP ewe, av}
for all places on the characteristic curve, and for every direction X and pw
through each place distinct from the tangential direction, shall have the same
sign: the sign being positive for a minimum, and negative for a maximum.
Tt isa necessary test. It is a sufficient test, for the type of strong varia-
tion imposed. But that type of strong variation is only a selected type.215) EXAMPLES OF THE WEIERSTRASS TEST 329
Eaamples of the E-function as a test.
215. The Weierstrass test is simple, in its application to all integrals
which are of the form [ (2, y) de or, what is the equivalent, ate of the form
[ren ers ynk ae
For such integrals, the H-function is
Xn a
fread a fons at
jee ava
telfen shay sen wel
which is equal to
Fey) (L-Qa' + py),
because (A + w)t=1, (w+ y%)#=1. If x be the inclination of QP to the
tangent to the curve, so that x lies between 0 and 2m for any direction that
is non-tangential, we have
E=f(@, y) (1 — cos x).
If f (2, y) does not change its sign along the characteristic curve, then—
as the sign of f(x, y) may be taken positive—we have H always positive.
‘Therefore a minimum is admissible, and does occur, so far as concerns the
requirement of strong variations of the selected type.
Es, 1. ‘Thus the testis satisfied for a catenoid of revolution. “The integral ix
ae fae;
and 20, a8 the tests are satisfied—with a limited rango—for weak vati
we still have a minimum,
Ex. 2. Similarly, the testis satisfied for a gravity brachistookrone, ‘The integral effect
ively (§ 34, Ex. 3) is
ions (§ 30, Ex. 1),
fo ta) bas,
‘where the positive sign is given to the radical.
‘The Weierstrass test, established (as above) for the case when PQ is a
straight line, can be extended to all integrals
[rends=[reyerrynta,
when P and Q are joined by any are (and not alone by a straight line).
As before, we have
iF
Taso Lan « (Er +030)
=f ey) Ey
Grey ot
Fe») Oat + wy’)
If (@ y) 008 x,330 WEIERSTRASS TEST SATISFIED BY GEODESICS [ou vir
where 1 is the length of the straight line PQ. For Jap, we have
Top= [f%, Ya,
where X, Y are the coordinates of any point on the are PQ as now drawn.
Hence, if the total length of the arc be L, we have
Ter= If, P),
where f(X, Y) represents a mean value of f(X, Y) along the are, while X
differs from 2, and Y differs from y, by quantities of the same order as Z at
least. Thus
FE, Y=f@ y+ (1,
where L represents the aggregate of the terms in Z, I’, ...; and so
Tap = Lf (@, y) + L(L}
=f (2,9) +(e],
that is, to the first order of small quantities,
= If sy).
Hence, to the first order of small quantities,
Lana + Toe - Laue =(L — boos x) f(a, y).
Now L is always greater than J, the rectilinear distanee between P and Q;
and therefore L — cos x is positive for all values of x. Consequently
Taxa Tor — Tours
for the integrals in question, is always positive, whatever form the continuous
are PQ may take, Hence the Weierstrass test, with the less specialised type
of aro, is satisfied for integrals [/(e, y) ds, thus admitting a minimum.
Ex. 3. The Weierstrass testis satisfied for all geodesics,
With the customary notation of differential geometry*, and when p, gare the parameters,
on a surface, any are on a surface is
freees28 pat ogres
and the geodesic quality is determined by the charactaristicequationt. For the eonstruction
of the Weierstms test, we take a point P given by p, q,and a neighbouring point @ given
by p-ted, q-+07; and then, asin the text, we have
Lam Taree {0 BB py Bet
(Bpe+2Fpg+ Gq" Ept+2Fp.g + Ga
pt
we {e(0%f +4) +9(rlE+0%)h.
Now if 4 %2e th spcecortintas Pandy Fy th of
X-Xametont(e} Ni- Yenttyntl’) A-Zeattant(e]s
* See my Lectures on Diferentiat Geometry, cbep. i.
+ 1b. chap. v.215] EXAMPLES 381
snd therefore
. Pa E+ 2Rby + Gr,
ee dr_, dp wy y iq
Goagend. E wih goaded
Thos
ann n$e-o%. Beer (647 2) + 0nd.
But the left-hand side
=leos yy
hor isthe length PQ, and ¥ isthe angle betweon PQ and the tangent tothe character.
istic ourve, If then, as before, x ix the angle between @P and this tangent, x=r 4;
— Lay Lane —1 008 y.
1 therefore the Weierstrass fancton is
1(1~co8y),
‘hich in always positive, ‘The Weierstrass tot is satis.
Also Jpp=U5
Ez. 4. Discuss the possible minimum of the integral
Se
taken between two given values of
‘The characteristic equation is
4 (ay
zd ® =0,
0 that the primitive is
yaartb,
where a and & are constants determined by the end-points (#,, 91) and (¢,, 93).
‘The Legendre tost gives 54, =2; itn ants fora minimum. As the characteristic
curve is a straight line, there is no finite conjugate of any initial point ; 0 the Jacobi
teat i satisfied unconditionally.
(@) When the integral is taken in the Weierstrass form, it is
Either of the (equivalent) characteristic equations leads to
yrartb,
For the Legendre test, we take eo
ap yt a”
which is positive if x i positive, There is no conjugate of a value &; for if
sao, ymacttb,
With the notation of §§ 61, 63, we have
HO=0, HO=0 W=ch Wi=1 smo ¥ Omac,
x=c e@=4
and the function Z(t, f9)=68(¢—%), which has no zero except f. Thus for general weak
variations, the integral admits a minimum,
and 20332 EXAMPLES NOT SATISFYING (ou. vi
But the subject of integration is rational (and so the Weierstrass test cannot be
satisfied : soo § 217). The Z-function is
=( (-Be)er( (2f-24)
aero,
‘and so changes sign with 2. ‘The straight line does not providea true minimum, that is, «
minimum also under strong variations,
(ii) We here have taken the change of sign of X as dominating the Z-funetion; and it
is sufficient, for our purpose, to determine whether H can change its sign. In this case,
other curves, not regular in direction but jagged, can easily be constructed which lead to
values smaller than the minimum under weak: variations.
Let AB be the straight line; PQ any element. Take PZ’ parallel to the axis of x,
where P7’ may be large compared with PQ, if we
please; and let R7Q=«, RPO: .
When for the portion PQ of AB, we sub- é
state the boken line PZT, along PO we have —
a
Wao and along 79 we have tan e; thus
the value of the intogral for P77¢
0, for PT; + TRtan® «, for 7Q; =TRtante -
QRtane
=detanatane.
‘The value of the element of the integral for PQ along AB is
detente;
8 ¢ ‘Thus both
the Legendre test and the Jacobi test admit the possibility of a minimum (necessarily for
‘weak variations alone, because only weak variations have been imposed) if , beginning at
1
48 value go, does not attain a value
%
(v) It therefore is desirable to trace the curve, We have
ae
Gr pora-w, Po -Za+na-wy;
‘and therefore a cusp (or some higher singularity) may be expected at g=—1.. We take
1
Inte
where » is small in the vicinity of the place. We find, for the values of and y in that
vicinity,
-e 2H (te wetfiee..),
‘showing that the place C is a cusp, the tangential direction at which is given by
an i
Ya 1. The two branches of the cure, op sitber |g
Sis af Un cup ar sven: C4 by a postive vaun of .
tad Oe ys aan Re gle OO
sear :
tng 04, ¢ wich =-{f) incre, 2 tat gs © iy
ese ae
ion ee
a
yowbotetblons) +e (set + wt
ae
7+ 1ep) 89h
fa
hence _
Along C4, ¢ is positive and g, is positive; while along CB, q ia positive and g, is negative,
‘Thus along C4, the quantity P is positive ; the Legendre test admits a rainimum for the
branch CA. And clearly ¢ is a negative constant.
‘Thus, for weak variations, the characteristic curve is
G04 y=wtalogg- a1,344 SOLID OF REVOLUTION (ox. vir
on changing the negative ¢ into a positive a. ‘The Legendre test is satisfied. ‘The Jacobi
test imposes no limit on a range, that begins anywhere on C/A and extends in the direction
Ctowards 4. Hence we have a minimut for weak variations,
(vi) But the Weierstrass test, for the simplest strong variation, was proved to be not
satisfied. What is a minimum, under weak variations, is not a minimum under the
simplest strong variation, ‘Therefore it is not a true minimum.
(Vii) Te was stated (§ 207) that Legendre obtained, by means of jagged lines a curve
(for the generation of the solid of revolution) which leads to an integral less than the
quasi-minimum provided by the characteristic curve for weak variations.
‘The following is « simplified form of Legendre’s construction* for the purpose. Let
PQ be any arc of the characteristic curve. Let QT be
‘the tangent at the lower extremity; through P, draw PC .
‘making with the axis of « the angle y equal to 70 ; and
through @ draw the ordinate VQC.
Along the arc PQ, we have
pi