2 JAI,
5.2 Linearly independent hyperplanes
Let n > 2, and let V be an n-dimensional Euclidean space with inner product
( , ). Let h be a nonzero vector in V. and let y E R. Define H, H(+'), and H(-') as follows:
H - IV EV I (h,v)-y}
H(+')
-
IV E V I (h, v) > y)H(-') - IV EV I (h,v)<y}.
The sets H(+') and H(-') are, respectively, the upper and lower open half-spaces determined by H. The vector h is called a normal vector to the hyperplane H. If 0 E H. then H is an (n - 1)-dimensional subspace of V. Note that 0 E H if and only if y - 0.
The set K in R" is convex if a, b E K implies that to + (1 - t)b E K for all t E [0, 1 ]. The sets H, H("), and H(-') are convex. For any subset Sofa Euclidean space V, the convex hull of S, denoted conv(S), is the smallest convex subset of V that contains S. Since the intersection of convex sets is convex, it follows that the convex hull of S is the intersection of all convex sets containing S. This intersection is nonempty since the Euclidean space V is convex and contains S.
Let H1, ... , H. be hyperplanes, and let H' - U 1 Hi. Let (1, -1 }m denote
the set of all m-tuples (µ 1 , ... , µm) such that µi E ( 1 , -1 ) f o r i -1, ...
, m.For(Iz1,...
E (1, -1 }m, let-n (w
mµm) I Hi
i-
The 2m sets H(µ1, ..., µm) are pairwise disjoint, and
V \ H' - U
H(Is1, .... Am).(N... R.)EI1.-l I'
Let V be an n-dimensional Euclidean space, and let Hl , ... , H be hyperplanes with normal vectors h 1 , ... , hm, respectively. The hyperplanes H1, ... , H,,, are linearly independent if the vectors h1, ... , hm are linearly independent. The hy- perplanes Hl , ... , Hm are linearly dependent if the vectors h i , ... , h", are linearly dependent.
136 5. Sums of vectors in Euclidean space
Lemma 5.1 Let Hl , ... , Hbe hyperplanes in an n-dimensional Euclidean space V. Suppose that 0 E H; for all i - 1, ... , m. The hyperplanes Hl , ... , Hm are linearly independent if and only if
H(E.1.1,...,µm)710 for all (µ1, ..., µm) E {1, -I }m.
Proof. There exist nonzero vectors h1, ... ,in V such that
Hi-IVEVI(hi,v)-0}
for i - I, ... , m. Suppose that H1, ... , Hm are linearly independent hyperplanes.
Then the vectors h 1, ... , hm are independent, and there exists a dual set of vectors
h ,
.. . , h,*, suchthat(hr,hi)-s,. - 0ifi7(j l if i - j
fori, j - 1,...,m.Let(µ1,...,{Lm) E (1, -1)'. Let
m
v - Eµih! E V.
I-1 Then
and so
m
(hi, v)->µi(hi.hi)-tLi
i-1
V E 1
for all i - I, ... , m. It follows that N(µ1, ... µ) 710.
Suppose that H1, ... Hm are linearly dependent hyperplanes. Then the vectors h 1, ... , hm are dependent, and there exist scalars a1, ... , am not all zero such that
E_1a;h;-0. Define
+1
if ai > 0
µi(
-1
if ai < 0.Then µiai > 0 for all i - and µiai > 0 for some j.
We shall show that H(µ1, ... , µm) - 0. If not, choose v E H(µ1, ... , µm).
Then V E H,1"'1 implies that
ai(hi, v) > 0
for i - 1, ..., m, and
It follows that
ai(hi,v) > 0.
m
0- (0, v)- (aaht.v -Eai(hi,v)>0,
r-1 .-1
which is impossible. This proves that N(µ1, ... , µm) - 0.
Lemma 5.2 Let V be an n-dimensional Euclidean space, and let Hl,
..
Hr be linearly independent hyperplanes in V such that 0 E H , f o r i - 1, ... , r. Thenr
dim
(flH); -n-r.
In particular, if r - n, then
n
n Hi - (0).
Proof. Let h, , ... , hr be normal vectors for the hyperplanes Hl, ... H, respec- tively. Then the set of vectors (h, , ... , hr ) is linearly independent, and (hi, v) - 0 for all v E H;. Let w - f;_, Hi, and let
Wl -{vE V I(v,w)-0forallwE W}.
Since h; E W -L for i - l , ... , r, it follows that dim(W ) > r, and so
dim(W) - n - dim(W') < n - r.
We shall prove that dim(W) > n - r by induction on r. If r - 1, then W - H, and dim(W) - dim(HI) - n - 1. Let 2 < r < n, and assume that the assertion is
true for r - 1. Let W' -n
it Hi. Then dim(W') > n - r + 1. Since W - W' n Hr, it follows thatdim(W)
-
dim(W' n HH)-
dim(W') + dim (Hr) - dim (W' + Hr)> dim(W') + dim (Hr) - dim V
> (n-r+I)+(n-1)-n
=
n-r.
Therefore, dim(W)-n-r.Ifr-n, then dim(f =, H;)-0,andsof;_i Hi =(0}.
Lemma 5.3 Let V be an n-dimensional Euclidean space and let HI .., H,,, be hyperplanes in V such that 0 E H; for i - 1, ..., m. Let S be a subset of V such that
SnH(µ,,...,k,,,)-/ 0
(5.1)forall(tl,,...,µ,,,)E (1,-1)m.Then
conv(S) n ; ,'0.
Proof. By induction on m. Let m - 1, and let h, be a normal vector for the hyperplane Hl. By (5.1), there exist vectors s, E S n Hi+,, and S2 E S n HI-1 such that
(h1,s,)-oil > 0
138 5. Sums of vectors in Euclidean space and
(h i, s2) - -
ThenS - 1 I s) +
(
a) +a2
I S2 E conv(S).
Since
- ,a2 - (11 C12 (h" S) - - 0,
a2(hI,s))+ac(h1,s2) a
a ,+a2 a,+a2 a ,+a2
it follows that $ E HI, and so
conv(S)fl H) 1 0.
Let m > 2, and suppose that the lemma holds for m - 1. Define S(+)) and S(-)) by
S
n
H,(.- Then (5.1) implies thatThen
S(+))n (n Hcµ,)m-i
-SnH(µ),...,IA.-,.+1)710
r-,
and m-I
s(- 1) n H(N,)) S f H(FLI, ... , lam-c, -1)10
f o r all (JAI,
km-1) E (1, -1
}m-). By the induction hypothesis, the lemma holds for the m - I hyperplanes H,, ... , Hm-1, and som-I
cony (S(+')) n n H;) 10.
This means that there exist vectors s) , ... , sk E S(+') and scalars a1,... ak E R
with a; > 0 for i - I... k and a, + ... + ak - I such that
k m-1 -1
s(+I) - E a; s; E conv(S('))) n n H; c_ con(s) n
m
H;
Moreover, s(+) E H.(+') since Si E
for i - I... k and
is convex.Similarly, there existss
a2 < 0.
m-) n,-I
s(-1) E cony (s(-1)) n n H;
con(s) n (n
H; I ,ands(-') E
Let T - {s(''), s(-0). Then
m-I 1
T c con(s) n n Hit ,
and so
m-i
conv(T) c con(s) n n Hi
.i-i
Since T n f 0 for µm E { 1, -1 }, it follows that there exists
s c- conv(T) n H. c con(s) n (Hi).
This completes the proof.
Lemma 5.4 Let V be an n-dimensional Euclidean space and let H1, ..., H be hyperplanes in V such that 0 E H, for all i - 1, ... , n. Let S be a subset of V such that
SnH(µi,...,µn)-SnlnH,ci)
// 0i- for all
µn) E 11, -1)". Then0 E conv(S).
Proof. It follows from Lemma 5.1 that the hyperplanes H1, ..., H,, are linearly independent and so, by Lemmas 5.2 and 5.3,
n
conv(S) n {0} - conv(S) n
(n Hil
0.f ) /
Lemma 5.5 Let V be an n-dimensional vector space, and let H1... H be lin- early independent hyperplanes in V with normal vectors h 1 , ... , h,,, respectively, and with 0 E Hi for all i - 1, ... , n. Let
n
Lj -nHi
;ri
forj-l....,n.Then
V - Hj®Lj
for j - 1, ... , n. Moreover, there exists a dual basis {h,, ... , h,, } for V such that
I
ifi-j
(hi, hi) - ai.j 0
if i 7Q
and Hj is the (n - 1)-dimensional subspace spanned by h,,...,hj_t, h!+,, ...
,hn and L j is the one-dimensional subspace spanned by h*.
140 S. Sums of vectors in Euclidean space
Proof. It follows from Lemma 5.2 that dim(Lj) = 1. Let fj* be a basis vector
for L. Then fi* E Lj implies that ff E H, for all i ¢ j, and so (hi, f'.) = 0 for
i j. Moreover,
(hj,fj)=0
if and only if
fj* EHjnLj=nH,=(0},
i_1
which is impossible since fJ 0. Therefore, (hj, 0 and
h' fi
ELj\Hj.
(h,, f;*)
Then (hi, h*) = Si,j for i, j = 1, ..., n, and Lj is spanned by h*. The vectors /t" are linearly independent, since _, xjh* = 0 implies that
0=(h,,0)=(hi,xjhf)xj(hi,h*)=xi
Itj-1 l-1
for i = 1, ..., n. Moreover, (hi, h*) = 0 for all i j implies that h, E Hj for i f j. Since Hj is a vector subspace of dimension n - 1, it follows that the set {hi,...,h*_i,h,... h) is a basis for H,, and V H
L. This completes the proof.Lemma 5.6 Let V be an n -dimensional vector space, and let H1, ... , H be lin- early independent hyperplanes in V with 0 E H i f o r i = 1, ... , n. Let Qi = Hi n H,, f o r i = 1 , ... , n - 1 . Then dim Qi = n - 2, and Q1 ,
pendent hyperplanes in H,,. Let
Q are linearly inde-
-i
L,, =nH1,
i_1
and let 7r : V -* H be the projection corresponding to the direct sum decompo- sition
V = H,, ® L,.
Let (tt 1, ... , E (1, - I and let V E V. If
,1 1
uEnHi')
then
,r(u) E n Qlu,)
i_i
Let S be a subset of V, and let rr(S) - (7r (s) S E S} c H,,. If
Sn H,")
0for all (µ,. ....
E (1, -l, then 0 E Conv(ir(S)).There exists a basis vector h,* for L,, such that if S C then ah, E Conv(S)
for some a > 0.
Proof. It follows from Lemma 5.2 that dim(Q;) - n - 2, and so Qi is a hyper-
plane in H for i - 1, ... , n - 1.
Let h, , ... , h be normal vectors f o r H, , ... , H,,, respectively, and let (h,, , h; )
qi
-
h,(h,
Since 1:1, ... , h,, are linearly independent vectors, it follows that the vectors
q,, ... ,
are linearly independent. Moreover, (h,,, qi) - 0, and so qi E H,,fori-1,...,n-1.
Letw E H,,.Then (gi,w)-(hi,w),and so w E Qi if and only if(gi,w)-0.
Thus, qi is a normal vector for Qi in the vector subspace H,,, and the hyperplanes
Q,, ... ,
are independent in H,,.By Lemma 5.5, there is a basis vector h, for L,, such that (hi, h,) - 8i,,,. If v E V, then
v - 7r (v) +rp(v)hn*, where n(v) E H and ip(v) E R. Moreover,
(qi, 7r (v))
- (hi,'r(v)) - (h, h,,) (ha, r(v))
_ (hi, 7r(v))
(hi, v) - rp(v)(h;, h,*) (hi, v)
for i - 1, ... , n - 1. It follows that
,1-I
VEnN(`')
i-i
if and only if
7r (V) E Q0.,).
ii
142 5. Sums of vectors in Euclidean space
If
-i
Sn
n
nHw,>
710 i-1
for all (µ i , E ( 1, -1)"-', then
n-1
7r (s) n
(
QcUO) f g
J
gfor a11(µ E (1, Lemma 5.4 implies that
0 E convOr(S)).
This means that there exist vectors sl, ... , sk E S with the property that
f o r some nonnegative scalars a,, ... , ak such that a, +... + ak - 1. Let si -7r(si)+(p(si)hn
and
Then
a - E amsi )- i-
k k k
aisi - a;7r(s;)+a;p(s;)h, -ahn E conv(S).
If S c
then(h s;) - (h,, , 7r (s;)) + rp(s;)(h
h,) - rp(s,) > 0
for all i - 1, ..., k, and so a > 0 and ahn E conv(S). This completes the proof.