Introduction
The follm·Ting 1:-asic ::-e~ult \-Taf::
proved
ipdepcndent.ly by P>-lmgren ( 1) ar:.d Calabi (2): ~"'l:Y :a-1:;1phero which isminimally
iP.mersed in thestandard 3-sphere
~s~be
an e~uat.or.This is an analogue
insphe- :r:ical
geometryof the fa,.-nous Bernstein theore.r:1 on rr.i.nimal hyper-
surfacesin three-dimensional
Euclideangeometry. After
thesolu-
t.ion of thE;! Bernst~in probler.t in Euclideangeometr..t
of 'higher~ ' . (d G' ' "1 .... . B b' ' ~"'' ""' ') ....,..
a.1mensJ.ons _ e ~J.Ct'9l.• .f'.-r.v:;;re:n, .,:)l.'7!0ns, om ;.erl., '-=~us
....
1. , ~..,;uernraised the question of
generalizing this result to higher-dimcn-
sional spheric~lgeometry (3). Tne following
easy(i.e. homogene-
ous) example of a minimalirnmersion
ofs
3into the standard
4.,..sphere S
4· (1)
wa~ 'known (4}; Let SO ( 3)
operate onthe symmetric (3x3)-matrices of trace zero
byconjugation, let s
4be the in-
variant uni,t
spher~. The principal orbit:is of type
S0(3)/z2
+ z2, so the ope of maxima+ .
vol~e .defines a non-equatorial minimal
immersion of s3 into
S~(1). The spherical Bernstein
problem WqStherefore formulated as a question about the existence of non- equatorial empeddingl$ of hyperspheres; there has been much recent progress
on th~s (5), (6), (10).However, the isolated nature
ofthe above counterexample in s
4(1 ), (it is the only such example which is homogeneous), still leaves open the natural question: To
which e;<tent ooes. the Almgren-Calabi theoremf.;1il in
highe;r dimen..-sions?- in view of the significance of this
forthe study
ofsin- gularities of min;i.lnc:tl
nypers~1rfacesit is a basic proolem
inRie- mannian geometry. !n t'llis paper we give the complete solution by
proving the
following theorem:
T'l;eorem 1.
be the EucliGean t111.i t. sphere of dir:1ension n, n)4., Then there exist infinitely many non-congruen-t minimal immersions of
s
n-1'ITiese sp11eres ere invar:~2.nt under a s il7lple .f~rans forrnation group 4>
The reJ.evant rr.ethcds f:::-om equivaria.::-1~.:. differer.tial georaetr.f to analyze the reduced minimal equation
in
orb:.t space, a sphericallune of angle ~, have been discussed in dctc.il in (6), (1 0); \ve sum!i'.arize the results in section 1. The existence of mini.TTI.al
hyperspheres is reduced to the existence
of
solution curves of the reduced equation i.vhich oscilla:te between two points on the singu- lar boundaries. This is established at one stroke :=or all dinten- sions by a technique of (1 0), which analyzes t11e variation of the pattern of critical points of auxilary funcitons along solution curves which e.vnanate £:-om the singular boundar.t of the spherical lune; this is discussed Section 2. This nethod requires less exp- li.cit computation than (6), (1 O); the fe.v necessary estimates and the proof of the main theorem are completed in Section 3.The important question of stability of the vertex singularity of the cones on these hypersphe=es may be cheCked by Simons'
criterion: If the norm of their second fundamental form is uni- forrnly bounded by ~, ~ne n-2 .... ~ cone is st~ble in Rn+l (8).
By
compu- ter assisted appro:dmations i t can be sho....rn that in many concrete cases, stability does in fact occur {11).1, Orbital geometry_and the reduced minimal eauation.
,.. ~e~ · G -
=
0 { ) p X 0 ( ) q ac~ · on :-:p+q ·.-~ - ~p ~ e. '""' .. · ~ c_. ·""'-a R 10Y
~ne • ' represen- t;;:tim'! ppe
pq 1.3e,
pp = standard repres,.7!ntation o£ 0 ( p) ,e ·-
trivial representation, {{:c,y,::;) E R' x R+x R}, defines the orbit space
!l :::::
the
p q. The space of ir-variants
is
additional equation ~.,
'J::: .c::. v~
...
~-- ;::::of the restriction a= this ~ction to
1 •
n ' $
s ;
nence is a spherical lune of angle2·
11: Ir:.t=ccqcing spher- ical polar coordi:1ates (r, e) E [0, itJ
x [0. ~L
wo have= Theorbital distance metric ds2
=
dr2 + sin2rde
2 • ~~e volume func-· · · h 1 f · · , b · · o+q-2
t1on, reg1ster1ng t e vo ume o pr1nc1pa~ or 1ts: v = s1n~ r s inp-l
e
cos q-le.
From standard formulas the mean curva '.: ure of a hypersurface represented by the curve y(s) ;:::: (r(s), e(s)) ir..orbit space, parameterized by a.rc ler:gth is H(y(s)) == J.:(y(s))-
~
ln v(y(s)) where k is the geodesic curvature and n is the anunit normal of the curve. By straightforward computation i t fol,- lO\V's that the hyper surface is minir:~al iff i t.s generating curve satisfies the differential equation:
(*) ..
r
=
cos a..
sinae =
sinr..
sin a cos aa
=
-(p+q-1) sinr cosr + sinrKe.
where K9
=
(p-1)cote -, (q-1)tane, and a is the angle from to the tangent vector.~nis equation becomes singular at the boundaries
e=O
andWe have the following main reduction theorem:
Theorem 2.
0
or
e
=2·
itLet y(s) = (r(s),
e(s)),
s E t(a,b) ·be a smooth curve in int X,pc.rametc:z:i::::ed by arc length, 'tlhich satisfies ( *). Assume tnat.
r(b-) E (O,n), e(a+)::::O, e(b-):::: ;. a:(a+)
=;
mod(2n),a (b-)= ~ (mod2 rt). Then y ( s) , sE [a, b] is the projection to X of a smooth, minimally L~mersed hypersphere of sn.
·ne observe that the condi·tion that y(s) enter the boundary orthogonally i•·nplies that the in·v·erse image is a smooth hypersur- face in Sn, its topological type is determined as the union of two napping cylinders det:ernined by the orbital data along ·r{ s) • For more details, see (1 0) .
We need more precise information about
(*)
at the singula;r boun- dary. ~IJe summarize:Proposition 1.
Let (b,O), bE(O,n), be any point in the interior of the singular boundary 9=0. Then there {s a unique analytic solution curve yb(s)
-
(rb(s), eb ( s)) of ( *) with rb(O) = b,eb(o)
=o.
Here c. ( 0+) nand (rb (sL " ( \ ~ {s)) depends analytically
=
2i tlb s} , onD
( s 'b) as long as s is restricted to an interval where yb(s) does not intersect the singular boundary again. A similar result holds for
e
= 'It 2"Proof:
This is proved by formal power series substitution and majoriza- tion. See (7).
Proposition 2.
Let bE(O, r.). J16ny solution curve (r{s), B(s)) of (*) which approaches (b,O) as s + 0-, must be analytic and hence coincide with the above unique solution (reversed in s).
/
.-·.
Proof:
It. is not hard to shO'Itl thc.t c:{O-)
= - -
)i; 2 (see (10)), hence the l i f t t.o is a smooth, r:1.inirnc:.l hypersuriace; i t follo<.vs from st.andard regularity theorews t.hat i t is analytic.The equation { *) is sym&;;etric
tn
respect to reflection of para- rr:cte:::-; hence any solution curve \.:hich hi t.s the boundary 1 can be continued back along the s~~e trajectory with a discontinuous jump J..n at the boundary; hence -' ci..L,
a!. solution curves may be conside-as defined for all s. Closeby solution curves will generically avoid the boundary, but we have the phenomenon of "sharp turning"
close to the boundary:
Proposition 3.
Let bE(O,~) and g>O. There exists a 6>0 such that for any solution curve y(s) = (r(s), e{s)) with r(O) = b, 9(0) <
o,
there is an s 0 E (O,c:) pectively ; - 9(0) <
&,
with 2!. +
2
8 (So ) < € 1
cr. (
r;o
>I
< c:(res- at the other bounda~J).
Here
the detailsbecome a
delicate estimate, see(10) for
thisargument:..
Proposition 4.
Special solution curves of
(*)
are: (a) b)c) sin2r cos2
e
=g-1
p+l-1 d) sin2r sin 2
e
= p+q-1 o-1Here (a) corresponds to t.he equator z=O of
sn.
The "meridian solution" (b) is the suspension of the principal orbitof
maximal volume, and does·not l i f t to a smooth hypersurface of· sp+q. The solutions (c) and (d) can be checked by careful computatiton, they:'.\
represent principal orbits of the larger sym:rr.etry groups 0 (p+1) x
O{q) and O(p) x O(q+l) on Sn, respectively.
Prooosit.ion 5.
Lcet ( r { s) ,
e (
s) 1 o: ( s) ) be any solution of ( *} • Then:( .:~) any relative maximurn (minimum) of
r(s)
occurs \•lith ->--~{ T'"(... 2 -
2) • n
(b) any relat.ive maximum ( rnin.imtun) of e(s) occurs with 6>90(6<90).
(c) any relative maximu.tn (ninimum) a(s) occurs v:it.h
in
t11e first or third (second or fourth) quadrant,
Proof:
(a) and (b) are immediate
by
differentiating (*)· Computi~g o:and subtituting the relation between
..
a, e, r defined by a = 0 yields:
a =
K ( r, e) sin a cos a: a·t a critical point of a:, wher~K(r1 9)
=
p+1-q-sin- 2 r[p+q-2+(p-1 )cot2e
+ (q-1) tan2e]is always negative2. Qualitative features of solution curves under deformations.
Definition 1 . Let the regions I-IV
in
orbit space be defined by:I: ( r, e) ,. n:
it)
(eo· !:) ( r,
e) ,.. (0 1 1t) (90, 1t'C: ( 2 8 X 2 . II: 0: 2 X
2).
III: ( r, e) E ( 0 1 2!)
2 X (0, eo) IV: ( r, e) E
<;.
it) X (0 1eo).
of ( *) with
r
b( 0) = (b u 0, ; ) e and (rb(s), eb(s)) to orbit space. Letbe the projection ri(b) be the i-th
r:l
maximum and minimum of rb(s), s>O, respectively, and define their
i i i i
corresponding arguments by rH(b)=rb(sMb), rm(b)
=
rb(smb). Theand Definition 3. Let
a!
(b) = a:.0 (tm~)
are defined simil<:J.rly.yb be the largest segment of + yb(s), s>O, vfl1ich does not touch the singular boundary.
::
Pronosition 6.
~~·---~~----
r: { )
Let. b E {0 1 2-). Then all critical points for rb s
along yb are non-degenerate and vary continuously with b. Ncne of them can coincide.
Proof:
By t.he uniqueness theorem for differential equations
l.
) ,
ri (b) -'--
",...
.,.
2 •-1'1 m From (*) \ve have
r
= {p+q-1 }cotr at a critical90ir:t of r( s). Hence these are all non-degenerate and stable;
frorn continuous dependence on initial conditions (Prop.l) i t fol- lows that i
5Hb'
s i
rnb vary continuously \vi th b. From Prop. 5 i t follc::t.rs that a critical point a is non-degenerate unless
s cosa = 0. Ass~~e ~{s
1
) = 0, cos~{s1
) = 0; from(*)
we havecosrb(s 1 )
=
0, i.e. rb(s 1 )= ;,
contradicting uniqueness of the equator solution. This argument also proves that critical points of rb(s) and ~0
(s) cannot coincide. If ~(s1
)=
O, sin~(s1
)= Og i t follows that K8 (s 1 ) = 0, i.e. eb(s1 )
=
e0 s contradicting·the uniqueness of the meridian solution. Hence. all critical points on yb.
+
are non-degenerate and stableffq.e.d.
Definition
4.
The N-th (r,a)-pattern of yb, P (b), N is the finite sequence of symbols constructed as follows: we assign symbolsi i i i
in the order the critical points
r~(b),
r r m~ aH, c:: rn same .as
i i i .,..
until
rm (b), aM (b), am(b) occur along yb, s>O, 'lfle have passed the N-th critical point or Yb (s) has reached the boundary.
Note: PN(b) is well-defined
~J
Proposition 6.Corollary.
AssumE: that +
y b has at least N i
critical points among the rN(b) 1
is locally constant around b=b1 . Proof:
....
By P::ropostion 1, yb' has at least N cri"'.:.ical p-oints for b
:::;uf:fic.iently close to b, . By P:=oposition 6 these must :wove conti- '
nuously c.nd cannot change: type under s~na.ll vc.:::iations of bi furthermorf.!e since they never coincide, they cc.nnot pass each other, and PN(b) is rigid
q.e.d.
Since any change in PN(b) must result from hitting the boundary, this i~ediately gives a criterion for establishing the
existence of "closed" solutions yb, oscillating bc.ck and forth between b and another boundary point. Let b E (06 ~) and
a: • <
0
at11: ~
a
= 2;
hence ~(s) < 0 for small positive s, ~ (s) >0 for smallD .
negative s, and c::b(s) may be considered to have a (not well- defined) maximum at s=O, coinciding \·lith a rnini."TTurn for rb ( s) •
s
irnilarly1 at an intersection IN"i th the boundary \ve have:In I: A rr~ximum for r(s) coincides \vith a maximum for a:(s).
In I I: A minimum for r { s) co inc ides with a minimum for a: ( s) • In IV: A :ruaximum for r( s) coincides with a minimum for a:(s).
We need to examine what happens under small perturbatons of the intersection situation.
Prooosition 7.
Let yb ( s) intersect the bounda::y at s=s 1 fer b=b1 E ( 0, ; ) . Variation of b around b
1 then all~1s a continuous crossing of critical points of a:b(s) and rb(s) at the intersection point according to the following scheme:
"."
.-J
·~,
Intersection pcint in
I: An rM(b)c:-osses
an a., (b) •J.."'l
I I : " 1\.,'1"1
.
r m (b) cress-es an am~ ,. o •)
III: An
r
m {b) c::osses an a:H(b).IV:
A.."'!. r ~1(b) crossesan
a: (b) • :mProof:
-
.
( ),.,
( )Assume ca.se IV,
l.~e.r 0 s 1
>2' 8b s 1 .
=0 By Pro- position
1:For any
c:>O thereis
a 6>0 and an s 0E(O,s 1 ) "lithe:, 0 < a. ( s0 )+ ; < g
D . - when
Jl..nalyzing
the sign of a:.. in combination
withProposi ton
5, yields the following possibilities:(i) reaches
a relativer:tinimum
a: J (b) > - n2. , by Proposi-m
..
tion
3i t then
turnssharply
upto
2 -7t e:. By estirnating
c:it reaches ; , .
(i.e.and
rM (b) ) iafter
a shorts-interval (which tends
to zerowith
6) •(ii}
decreases toas approaches an
intersec-tion
withthe
bounda~.decreases
past 'it2 (i.e.
and · r~(b))
and turnsh 1 3'lt h .
s arp
y
to - ~+
e.By
c oos~ng e: small,we
may ass~e that we remainin
!V. The a:b ( s)now
reaches a relativeminimum
aftel;"a a-small
(i.e.
decreasesto zero
with o)s-interval, otherwise,
an estimateof
a • showsthat
:::.-eacheswhile still in IV, contradicting Proposition S(a). Hence, in
case(i)
an a (b)m
precedes an
rMi(b) , and in(iii) and r~(b)
precedes and a.mj(b),
approaching
eachother as o~o. In case (ii) they coincide. This proves the propositon in region IV. The /
regionIII foll(JWs
by reflection aroundthe symmetry
linea similar
argumentat
the boundaryI
andII:
.,..
e =
2concludes the proof for
q.e.d.
3. Small perturbations: of solutions and oroof of the main theorem.
'D • t 8
,_ropes~ on .
Let r(s) =
(r(s),S{s),
~(s))be a
~elution of(x)
with0
< a ( 0) < ~,
a (
0) < 0, ( r ( 0) ,e (
0) ) E ! I I. Theny (
s}=
(:de) ,e(s)) escapes the region III by crossing 9 ;::: e0 fo::.- an s :::; s0
> 0; furthermore ~(s) < 0 for s E (0, s 0
j.
In partucular this holds for yb, b E (0, ~).Proof.
..
"In I I.I we have a > 0 at ~=0 and a < 0 at . c:
= ·:2u
':1: hence cx(s) E (0, ~)until y(s) escapes III.By
Prcpositon 5(c), c:(s) has no relative· minimum, henc::e ~(s) rerr.ains negative until the es-..
( ) 'it h
cape. By
*
we have a > 0 at r =2 ,
ence the escape must be across 9= 9 • Since<;
~ n) u we must have 0 follows.yb(s) starts in III and ~(s) < 0 for a E
·~(s) < ~· to begin with, and the result q.e.d.
We need to exacine the (r,a)-patterns of yb(s) close to the kncwn solutions b = ; and b
= ( P'~~~ 1 )~.
Since the case n=
p+ q < 9 of Theore.'n 1 is covered in (51 6), we assume n ;;. 9: for technical simplicity we also assu1ne, without loss of generality, that q - 1 .;;; p
<
q + 1 •Proposition 9.
With the above assumptions we have: For any N there exists
an
~2N ( ) ( 1 1 2 2 .,..N N )
such that P b = rM, rm, rH, rm, ••• , ""M' r~ for b~ ... 2 (.!£
-e:, 2 .
'It)Proof:
Let b E (
~-e:, ~)and
let yb(s) cross from III to II for· s=s1 • From (*) we have ~(s) < 0 for a E (0, ~} in II. By continuous dependence on initial conditions: For anyo
we maychoose e sosrnallthat lrb(s)-~1 < 5u IO:b(s) -~1 < 5 until eb < s 3 )
=
e 0 - 5 •Lernma.
For sufficiently s:rnall o, yb ( s) crosses r
= 2
1t before s~s3 •Proof:
Define the auxilary function Differentiation gives:
t: ( s) -- cos o:(s)
cos r(s)"
(1)
u
= sin-1r[(p+q-1 )sin2c, - usinai\0 + u2sinr]. By Propositionv
5 (a), cos c:(s) > 0 for r(s) < it 2' so u(s) > 0 and K 9 (s) < 0 for s E (sl,s2) as long as y- ( s) does not cross into
r.
HenceD
"
u > (p+q-l)sin2a: + u2 in this region, and, by choosing E small:
u
> (p+q-2) + u2. By comparison \vi th the solution u = (p+q-2)~ ltan ((p+q-2);2(s-s1 ) +C) of the equation (2)
u
= (p+q-2) +.u 2 , wathat u(~) blows infinity, i.e. yb(s) ':t
see up to crosses r =
2
before s =
.
51 + 'lt 2(p+q-2) -~ < 51 + 216' it and the Lemmafollows.
We na.ll' have ~b(s) < 0 for s E (O,s 2 ), '.vhere rb(s2 ) = ;, s2 E
(s
1 , s3 ). Furth~=more, by Proposition S(c), ~(s)..
< 0 alsofor
s E[s
2 , s3 ]. Combining this argument with.the sharp turning of"
Proposition 3 and a sig~ discussion for a from
(*),
we see that~ (s) decreases rapidly through 0 to ~(s4) :::::
- 2
1t +&.
Then u( s4 ) < 0, rb(s) .,. 2' 1t ab(s) <-2
1t until c:b(ss) =-2,
1t i.e. rM(b)= 1 rb(ss). By checking signs of (1) '"emay
compare with (2) as above, and conclude that while is still in I. By choosing e small we have: rb(s 5 ) - ~ <o.
By the argumentof
Proposition 8 yb(s) must now escape I by crossinge
=e
0 intoIV: By combining Propositions 1, 3, S(c), sign discussion for from
(*),
and the aux;ilary differential equation (1) as above,a •
we show: yb(s) must cross r
=
1t 2 on its way dm.;n, turn sharply near the boundary 9=0 of I!Io a continues to decrease1 •
rrn(b) = rb(s 6 ), ab(s) < 0 for s E (O,s 6 ]. Here
E-small, so we are in the same situation as we started with.
.
. -·~
'•
;
Hencet the argument may be =epeated N times to conclude the
proof of Propositon
9.q.e.d.
Propositon
10.For b
sufficiently close to Arcs.l.n( .
' q-1 1 )~ , P"' (b) ,p-rq-
is either
1) ( 1 . 1)
a m
or
a , :m.rM •
Proof: Straightforward computation shows
thatfor
thesolution
. q-1 ~2 ..
yb(s) with
b =ArcsJ.n(p+q;.,.l),,
ab(s) < 0until its first inter- section with the boundary of IV. By propositons 1 and 7 the
~esultfollows.
Proof of Theorem 1.
for
c E (b, ~)}and for
c E (b, ;)}.Byq-1 ~ 'it
(p+q-l ) ,
2 ). ByProposi- tion
7rc. and
Yb.intersect
theboundaries of
I Iand IV at
.J. l.
r! and ri respectivey (wee must 11ave P2i+1
(c) =
M
1 1
i
1ri)
(rM, rm,
• • • IrM,
a;m' m for
c E(ci-e:,
c.) l.for some
£) • ByTheore..tn
2the rc. represent an infinite sequence of generating
l.
curves for minimally immersed hyperspheres of
q.e.d.
Remark 1.
From propositions
9and
10it follows that the initial points
c .~
of
the generating curves
val [Arc in( q-l )~ n ]must accumulate somewhere in the inter-
5
p+q-1 , 2 .
We observe thatif
does notintersect the boundary
fors>O when
bvaries
insome interval
J,this would imply that PN(b) is constant for b in
Jfor
all N.
By refining
the abovetechniques it should
thenbe
possible to prove that yb lifts to a minimally immersed hyper- sphere fo~
a
dense set of b in ( 0, 1t) ,Remark 2
The Yb. Fepresent an infinite sequence of
miniw~llyimmersed
.l.
generalized Clifford
tori.References
1. F.J. Almgren, Jr.: Some interior
regu~aritytheorems for mini-
malsurfaces and
an extensionof
Bernstein's teorem.Ann. of Mathw
(2) 84 (1966), 277-292.2. E.
Calabi: Minimalimmersions of
surfacesin
Euclideanspheres
J. Diffe
geom.
1 (1967), ppe 111-125 •. 3. S. s. Chern: Differential
geometry,its past
andfuture. Actes, Congre
Intern. Math. Tome1 (1970),
pp.41-53.
4. w.Y. Hsian9:
Remarkson
closed minirna.lsubmanifolds
inthe
standard
rienaannian
m-sphere. J.Diff.
Geom. 1 (1967),pp.
257 267.5.
'11.Y. Hsiang: Minimal
conesand the spherical
Bernstein problem I.Annals of Math. 118
(1983), pp. 61-7~.6. W.Y. Hsiang: Minimal cones
andthe spherical Bernstein
problem II. Invent. Math. 74 (1983), pp. 351-369.7. W. T. Hsiang,
and W. Y.Hsiang:
On theexistence of
codim~m$ionone nrl:.nimal spheres in com,pact symmetric spaces of rank 2, II.
J •.