Problems from the Book Problem 19.

9
Let n N. Let w
1
, w
2
, ..., w
n
be n reals. Prove the inequality
n

i=1
n

j=1
ijw
i
w
j
i +j 1

_
n

i=1
w
i
_
2
.
Solution by Darij Grinberg
Notations.
For any matrix A, we denote by A
_
j
i
_
the entry in the j-th column and the
i-th row of A. [This is usually denoted by A
ij
or by A
i,j
.]
Let k be a eld. Let u N and v N, and let a
i,j
be an element of k for every
(i, j) {1, 2, ..., u}{1, 2, ..., v} . Then, we denote by (a
i,j
)
1jv
1iu
the uv matrix
A which satises A
_
j
i
_
= a
i,j
for every (i, j) {1, 2, ..., u} {1, 2, ..., v} .
Let n N. Let t
1
, t
2
, ..., t
n
be n objects. Let m {1, 2, ..., n}. Then, we let
_
t
1
, t
2
, ...,

t
m
, ..., t
n
_
denote the (n 1)-tuple (t
1
, t
2
, ..., t
m2
, t
m1
, t
m+1
, t
m+2
, ..., t
n
)
(that is, the (n 1)-tuple (s
1
, s
2
, ..., s
n1
) dened by s
i
=
_
t
i
, if i < m;
t
i+1
, if i m
for
all i {1, 2, ..., n 1}).
Let R be a commutative ring with unity. Let a
1
, a
2
, ..., a
m
be m elements of R.
Then, we dene an element
k
(a
1
, a
2
, ..., a
m
) of R by

k
(a
1
, a
2
, ..., a
m
) =

S{1,2,...,m};
|S|=k

iS
a (i) .
This element
k
(a
1
, a
2
, ..., a
m
) is simply the k-th elementary symmetric polyno-
mial evaluated at a
1
, a
2
, ..., a
m
.
The Viete theorem states that

{1,2,...,m}
(x a

) =
m

k=0
(1)
k

k
(a
1
, a
2
, ..., a
m
) x
mk
for every x R. If we choose some i {1, 2, ..., m} and apply this equality to
the m 1 elements a
1
, a
2
, ..., a
i
, ..., a
m
in lieu of the m elements a
1
, a
2
, ..., a
m
,
then we obtain

{1,2,...,m}\{i}
(x a

) =
m1

k=0
(1)
k

k
(a
1
, a
2
, ..., a
i
, ..., a
m
) x
m1k
. (1)
1
Theorem 1 (Sylvester). Let n N, and let A R
nn
be a symmetric
n n matrix. Then, the matrix A is positive denite if and only if every
m {1, 2, ..., n} satises det
_
_
A
_
j
i
__
1jm
1im
_
> 0.
For a proof of Theorem 1, see any book on symmetric or Hermitian matrices.
Theorem 2 (Cauchy determinant). Let k be a eld. Let m N. Let
a
1
, a
2
, ..., a
m
be m elements of k. Let b
1
, b
2
, ..., b
m
be m elements of k.
Assume that a
j
= b
i
for every (i, j) {1, 2, ..., m}
2
. Then,
det
_
_
1
a
j
b
i
_
1jm
1im
_
=

(i,j){1,2,...,m}
2
;
i>j
((a
i
a
j
) (b
j
b
i
))

(i,j){1,2,...,m}
2
(a
j
b
i
)
.
In the following, I attempt to give the most conceptual proof of Theorem 2. First
we recall a known fact we are not going to prove:
Theorem 3 (Vandermonde determinant). Let S be a commutative
ring with unity. Let m N. Let a
1
, a
2
, ..., a
m
be m elements of S. Then,
det
_
_
a
j1
i
_
1jm
1im
_
=

(i,j){1,2,...,m}
2
;
i>j
(a
i
a
j
) .
Besides, a trivial fact:
Lemma 4. Let S be a commutative ring with unity. Let a S. In the
ring S [X] (the polynomial ring over S in one indeterminate X), the element
X a is not a zero divisor.
And a consequence of this fact:
Lemma 5. Let R be a commutative ring with unity. Let m N. In the
ring R[X
1
, X
2
, ..., X
m
] (the polynomial ring over R in m indeterminates X
1
,
X
2
, ..., X
m
), the element

(i,j){1,2,...,m}
2
;
i>j
(X
i
X
j
) is not a zero divisor.
Proof of Lemma 5. We will rst show that:
For any (i, j) {1, 2, ..., m}
2
satisfying i > j, the element X
i
X
j
of the ring R[X
1
, X
2
, ..., X
m
] is not a zero divisor. (2)
Proof of (2). Let R
_
X
1
, X
2
, ...,

X
i
, ..., X
m
_
denote the sub-R-algebra of R[X
1
, X
2
, ..., X
m
]
generated by the elements X
1
, X
2
, ..., X
i2
, X
i1
, X
i+1
, X
i+2
, ..., X
m
(that is, the m ele-
ments X
1
, X
2
, ..., X
m
except of X
i
). Consider the ring
_
R
_
X
1
, X
2
, ...,

X
i
, ..., X
m
__
[X]
2
(this is the polynomial ring over the ring R
_
X
1
, X
2
, ...,

X
i
, ..., X
m
_
in one indeterminate
X). It is known that there exists an R-algebra isomorphism :
_
R
_
X
1
, X
2
, ...,

X
i
, ..., X
m
__
[X]
R[X
1
, X
2
, ..., X
m
] such that (X) = X
i
and (X
k
) = X
k
for every k {1, 2, ..., m} \
{i}. Hence, (X X
j
) = (X)
. .
=X
i
(X
j
)
. .
=X
j
, as
j{1,2,...,m}\{i}
= X
i
X
j
. Since X X
j
is not a zero
divisor in
_
R
_
X
1
, X
2
, ...,

X
i
, ..., X
m
__
[X] (by Lemma 4, applied to S = R
_
X
1
, X
2
, ...,

X
i
, ..., X
m
_
and a = X
j
), it thus follows that (X X
j
) = X
i
X
j
is not a zero divisor in
R[X
1
, X
2
, ..., X
m
] (since is an R-algebra isomorphism). This proves (2).
It is known that if we choose some elements of a ring such that each of these elements
is not a zero divisor, then the product of these elements is not a zero divisor. Hence,
(2) yields that the element

(i,j){1,2,...,m}
2
;
i>j
(X
i
X
j
) of the ring R[X
1
, X
2
, ..., X
m
] is not
a zero divisor. This proves Lemma 5.
Now comes a rather useful fact:
Theorem 6. Let R be a commutative ring with unity. Let m N.
Consider the ring R[X
1
, X
2
, ..., X
m
] (the polynomial ring over R in m in-
determinates X
1
, X
2
, ..., X
m
). Then,
det
_
_
(1)
mj

mj
_
X
1
, X
2
, ...,

X
i
, ..., X
m
__
1jm
1im
_
=

(i,j){1,2,...,m}
2
;
j>i
(X
i
X
j
) .
Proof of Theorem 6. Let V =
_
X
j1
i
_
1jm
1im
. Then, V
_
j
i
_
= X
j1
i
for every
i {1, 2, ..., m} and j {1, 2, ..., m}, and
det V = det
_
_
X
j1
i
_
1jm
1im
_
=

(i,j){1,2,...,m}
2
;
i>j
(X
i
X
j
) (3)
(by Theorem 3, applied to S = R[X
1
, X
2
, ..., X
m
] and a
i
= X
i
).
Let W =
_
(1)
mj

mj
_
X
1
, X
2
, ...,

X
i
, ..., X
m
__
1jm
1im
. Then,
W
_
j
i
_
= (1)
mj

mj
_
X
1
, X
2
, ...,

X
i
, ..., X
m
_
for every i {1, 2, ..., m} and j
{1, 2, ..., m}.
For every i {1, 2, ..., m} and j {1, 2, ..., m}, we can apply (1) to x = X
j
and
a
k
= X
k
, and obtain

{1,2,...,m}\{i}
(X
j
X

) =
m1

k=0
(1)
k

k
_
X
1
, X
2
, ...,

X
i
, ..., X
m
_
X
m1k
j
. (4)
Now, for every i {1, 2, ..., m} and j {1, 2, ..., m}, we have
3
_
WV
T
_
_
j
i
_
=
m

k=1
W
_
k
i
_
. .
=(1)
mk

mk(X
1
,X
2
,...,
c
X
i
,...,Xm)
V
T
_
j
k
_
. .
=V
2
4
k
j
3
5
=X
k1
j
=
m

k=1
(1)
mk

mk
_
X
1
, X
2
, ...,

X
i
, ..., X
m
_
X
k1
j
=
m1

k=0
(1)
k

k
_
X
1
, X
2
, ...,

X
i
, ..., X
m
_
X
m1k
j
(here, we substituted k for mk in the sum)
=

{1,2,...,m}\{i}
(X
j
X

) (by (4)) . (5)

Thus, if j = i, then
_
WV
T
_
_
j
i
_
= 0 (since
_
WV
T
_
_
j
i
_
=

{1,2,...,m}\{i}
(X
j
X

) ,
but the product

{1,2,...,m}\{i}
(X
j
X

) contains the factor X

j
X
j
= 0 and thus
equals 0). Hence, the matrix WV
T
is diagonal. Therefore,
det
_
WV
T
_
=
m

i=1
_
WV
T
_
_
i
i
_
=
m

i=1

{1,2,...,m}\{i}
(X
i
X

)
_
_
since (5), applied to j = i, yields
_
WV
T
_
_
i
i
_
=

{1,2,...,m}\{i}
(X
i
X

)
_
_
=

(i,){1,2,...,m}
2
;
=i
(X
i
X
j
) =

(i,j){1,2,...,m}
2
;
j=i
(X
i
X
j
) =

(i,j){1,2,...,m}
2
;
j>i
(X
i
X
j
)

(i,j){1,2,...,m}
2
;
i>j
(X
i
X
j
)
_
since the set
_
(i, j) {1, 2, ..., m}
2
| j = i
_
is the union of the two disjoint sets
_
(i, j) {1, 2, ..., m}
2
| j > i
_
and
_
(i, j) {1, 2, ..., m}
2
| i > j
_
_
.
But on the other hand,
det
_
WV
T
_
= det W det
_
V
T
_
= det W

(i,j){1,2,...,m}
2
;
i>j
(X
i
X
j
)
(since det
_
V
T
_
= det V =

(i,j){1,2,...,m}
2
;
i>j
(X
i
X
j
)). Hence,
det W

(i,j){1,2,...,m}
2
;
i>j
(X
i
X
j
) = det
_
WV
T
_
=

(i,j){1,2,...,m}
2
;
j>i
(X
i
X
j
)

(i,j){1,2,...,m}
2
;
i>j
(X
i
X
j
) .
But since the element

(i,j){1,2,...,m}
2
;
i>j
(X
i
X
j
) of the ring R[X
1
, X
2
, ..., X
m
] is not a
4
zero divisor (according to Lemma 5), this yields
det W =

(i,j){1,2,...,m}
2
;
j>i
(X
i
X
j
) .
Since W =
_
(1)
mj

mj
_
X
1
, X
2
, ...,

X
i
, ..., X
m
__
1jm
1im
, this becomes
det
_
_
(1)
mj

mj
_
X
1
, X
2
, ...,

X
i
, ..., X
m
__
1jm
1im
_
=

(i,j){1,2,...,m}
2
;
j>i
(X
i
X
j
) .
Thus, Theorem 6 is proven.
Next, we show:
Theorem 7. Let R be a commutative ring with unity. Let m N. Let a
1
,
a
2
, ..., a
m
be m elements of R. Let b
1
, b
2
, ..., b
m
be m elements of R. Then,
det
_
_
_
_

{1,2,...,m}\{i}
(a
j
b

)
_
_
1jm
1im
_
_
=

(i,j){1,2,...,m}
2
;
i>j
((a
i
a
j
) (b
j
b
i
)) .
Proof of Theorem 7. Consider the ring R[X
1
, X
2
, ..., X
m
] (the polynomial ring over
R in m indeterminates X
1
, X
2
, ..., X
m
).
Let

V =
_
a
j1
i
_
1jm
1im
. Then,

V
_
j
i
_
= a
j1
i
for every i {1, 2, ..., m} and j
{1, 2, ..., m}.
Let W =
_
(1)
mj

mj
_
X
1
, X
2
, ...,

X
i
, ..., X
m
__
1jm
1im
. Then,
W
_
j
i
_
= (1)
mj

mj
_
X
1
, X
2
, ...,

X
i
, ..., X
m
_
for every i {1, 2, ..., m} and j
{1, 2, ..., m}.
For every i {1, 2, ..., m} and j {1, 2, ..., m}, we can apply (1) to x = a
j
and
a
k
= X
k
, and obtain

{1,2,...,m}\{i}
(a
j
X

) =
m1

k=0
(1)
k

k
_
X
1
, X
2
, ...,

X
i
, ..., X
m
_
a
m1k
j
. (6)
Now, for every i {1, 2, ..., m} and j {1, 2, ..., m}, we have
5
_
W

V
T
_
_
j
i
_
=
m

k=1
W
_
k
i
_
. .
=(1)
mk

mk(X
1
,X
2
,...,
c
X
i
,...,Xm)

V
T
_
j
k
_
. .
=
e
V
2
4
k
j
3
5
=a
k1
j
=
m

k=1
(1)
mk

mk
_
X
1
, X
2
, ...,

X
i
, ..., X
m
_
a
k1
j
=
m1

k=0
(1)
k

k
_
X
1
, X
2
, ...,

X
i
, ..., X
m
_
a
m1k
j
(here, we substituted k for mk in the sum)
=

{1,2,...,m}\{i}
(a
j
X

) (by (6)) .
Hence,
W

V
T
=
_
_

{1,2,...,m}\{i}
(a
j
X

)
_
_
1jm
1im
.
Thus,
det
_
_
_
_
_
_
_
_
_

{1,2,...,m}\{i}
(a
j
X

)
_
_
1jm
1im
. .
=W
e
V
T
_
_
_
_
_
_
_
= det
_
W

V
T
_
= det W det
_

V
T
_
=

(i,j){1,2,...,m}
2
;
j>i
(X
i
X
j
)

(i,j){1,2,...,m}
2
;
i>j
(a
i
a
j
)
_
_
_
_
_
_
_
_
_
since det W = det
_
_
(1)
mj

mj
_
X
1
, X
2
, ...,

X
i
, ..., X
m
__
1jm
1im
_
=

(i,j){1,2,...,m}
2
;
j>i
(X
i
X
j
)
by Theorem 6 and det
_

V
T
_
= det

V = det
_
_
a
j1
i
_
1jm
1im
_
=

(i,j){1,2,...,m}
2
;
i>j
(a
i
a
j
)
by Theorem 3
_
_
_
_
_
_
_
_
_
=

(j,i){1,2,...,m}
2
;
i>j
(X
j
X
i
)

(i,j){1,2,...,m}
2
;
i>j
(a
i
a
j
)
(here, we renamed i and j as j and i in the rst product)
=

(i,j){1,2,...,m}
2
;
i>j
(X
j
X
i
)

(i,j){1,2,...,m}
2
;
i>j
(a
i
a
j
)
=

(i,j){1,2,...,m}
2
;
i>j
((X
j
X
i
) (a
i
a
j
)) =

(i,j){1,2,...,m}
2
;
i>j
((a
i
a
j
) (X
j
X
i
)) .
6
Both sides of this identity are polynomials over the ring R in m indeterminates X
1
, X
2
,
..., X
m
. Evaluating these polynomials at X
1
= b
1
, X
2
= b
2
, ..., X
m
= b
m
, we obtain
det
_
_
_
_

{1,2,...,m}\{i}
(a
j
b

)
_
_
1jm
1im
_
_
=

(i,j){1,2,...,m}
2
;
i>j
((a
i
a
j
) (b
j
b
i
)) .
This proves Theorem 7.
Proof of Theorem 2. For every i {1, 2, ..., m} and j {1, 2, ..., m} , we have
1
a
j
b
i
=

{1,2,...,m}\{i}
(a
j
b

{1,2,...,m}
(a
j
b

)
.
Hence, the matrix
_
1
a
j
b
i
_
1jm
1im
is what we obtain if we take the matrix
_

{1,2,...,m}\{i}
(a
j
b

)
_
1jm
1im
and divide its j-th column by

{1,2,...,m}
(a
j
b

) for every j {1, 2, ..., m}. Therefore,

det
_
_
1
a
j
b
i
_
1jm
1im
_
=
det
_
_
_

{1,2,...,m}\{i}
(a
j
b

)
_
1jm
1im
_
_

j{1,2,...,m}

{1,2,...,m}
(a
j
b

)
=
det
_
_
_

{1,2,...,m}\{i}
(a
j
b

)
_
1jm
1im
_
_

(,j){1,2,...,m}
2
(a
j
b

)
=

(i,j){1,2,...,m}
2
;
i>j
((a
i
a
j
) (b
j
b
i
))

(i,j){1,2,...,m}
2
(a
j
b
i
)
(by Theorem 7). Thus, Theorem 2 is proven.
Theorem 8. Let n N. Let a
1
, a
2
, ..., a
n
be n pairwise distinct reals. Let
c be a real such that a
i
+ a
j
+ c > 0 for every (i, j) {1, 2, ..., n}
2
. Then,
the matrix
_
1
a
i
+a
j
+c
_
1jn
1in
R
nn
is positive denite.
Proof of Theorem 8. Let A =
_
1
a
i
+a
j
+c
_
1jn
1in
. Then, A
_
j
i
_
=
1
a
i
+a
j
+c
for every i {1, 2, ..., m} and j {1, 2, ..., m}.
Thus, A R
nn
is a symmetric n n matrix.
Dene n reals b
1
, b
2
, ..., b
n
by b
i
= a
i
c for every i {1, 2, ..., n} . Then, a
j
= b
i
for every (i, j) {1, 2, ..., n}
2
(since a
j
b
i
= a
j
(a
i
c) = a
i
+a
j
+c > 0).
7
Now, every m {1, 2, ..., n} satises
det
_
_
A
_
j
i
__
1jm
1im
_
= det
_
_
1
a
j
b
i
_
1jm
1im
_
_
since A
_
j
i
_
=
1
a
i
+a
j
+c
=
1
a
j
(a
i
c)
=
1
a
j
b
i
_
=

(i,j){1,2,...,m}
2
;
i>j
((a
i
a
j
) (b
j
b
i
))

(i,j){1,2,...,m}
2
(a
j
b
i
)
_
by Theorem 2, since a
j
= b
i
for every (i, j) {1, 2, ..., m}
2
_
=

(i,j){1,2,...,m}
2
;
i>j
(a
i
a
j
)
2

(i,j){1,2,...,m}
2
(a
i
+a
j
+c)
_
since (a
i
a
j
) (b
j
b
i
) = (a
i
a
j
) ((a
j
c) (a
i
c)) = (a
i
a
j
)
2
and a
j
b
i
= a
j
(a
i
c) = a
i
+a
j
+c
_
> 0
(since (a
i
a
j
)
2
> 0 for every (i, j) {1, 2, ..., m}
2
satisfying i > j (because a
1
, a
2
, ...,
a
n
are pairwise distinct, so that a
i
= a
j
, thus a
i
a
j
= 0), and a
i
+a
j
+c > 0 for every
(i, j) {1, 2, ..., m}
2
).
Hence, according to Theorem 1, the symmetric matrix A is positive denite. Since
A =
_
1
a
i
+a
j
+c
_
1jn
1in
, this means that the matrix
_
1
a
i
+a
j
+c
_
1jn
1in
is positive
denite. Thus, Theorem 8 is proven.
Corollary 9. Let n N. Let a
1
, a
2
, ..., a
n
be n pairwise distinct reals.
Let c be a real such that a
i
+a
j
+c > 0 for every (i, j) {1, 2, ..., n}
2
. Let
v
1
, v
2
, ..., v
n
be n reals. Then, the inequality
n

i=1
n

j=1
v
i
v
j
a
i
+a
j
+c
0 holds,
with equality if and only if v
1
= v
2
= ... = v
n
= 0.
Proof of Corollary 9. Dene a vector v R
n
by v =
_
_
_
_
v
1
v
2
...
v
n
_
_
_
_
. Then,
v
T
_
1
a
i
+a
j
+c
_
1jn
1in
v =
n

i=1
n

j=1
1
a
i
+a
j
+c
v
i
v
j
=
n

i=1
n

j=1
v
i
v
j
a
i
+a
j
+c
. (7)
Also, obviously,
v = 0 holds if and only if v
1
= v
2
= ... = v
n
= 0. (8)
Now, since the matrix
_
1
a
i
+a
j
+c
_
1jn
1in
R
nn
is positive denite (by Theorem
8), we have v
T
_
1
a
i
+a
j
+c
_
1jn
1in
v 0, with equality if and only if v = 0. According
8
to (7) and (8), this means that
n

i=1
n

j=1
v
i
v
j
a
i
+a
j
+c
0, with equality if and only if
v
1
= v
2
= ... = v
n
= 0. Thus, Corollary 9 is proven.
Corollary 10. Let n N. Let a
1
, a
2
, ..., a
n
be n pairwise distinct reals.
Let c be a real such that a
i
+ a
j
+ c > 0 for every (i, j) {1, 2, ..., n}
2
.
Let w
1
, w
2
, ..., w
n
be n reals. Then, the inequality
n

i=1
n

j=1
a
i
a
j
w
i
w
j
a
i
+a
j
+c

c
_
n

i=1
w
i
_
2
holds, with equality if and only if (c +a
1
) w
1
= (c +a
2
) w
2
=
... = (c +a
n
) w
n
= 0.
Proof of Corollary 10. Dene n reals v
1
, v
2
, ..., v
n
by v
i
= (c +a
i
) w
i
for every
i {1, 2, ..., n} .
Then,
n

i=1
n

j=1
a
i
a
j
w
i
w
j
a
i
+a
j
+c

_
_
c
_
n

i=1
w
i
_
2
_
_
=
n

i=1
n

j=1
a
i
a
j
w
i
w
j
a
i
+a
j
+c
+c
_
n

i=1
w
i
_
2
=
n

i=1
n

j=1
a
i
a
j
w
i
w
j
a
i
+a
j
+c
+c
n

i=1
n

j=1
w
i
w
j
_
_
since
_
n

i=1
w
i
_
2
=
n

i=1
n

j=1
w
i
w
j
_
_
=
n

i=1
n

j=1
_
a
i
a
j
w
i
w
j
a
i
+a
j
+c
+cw
i
w
j
_
=
n

i=1
n

j=1
_
a
i
a
j
a
i
+a
j
+c
+c
_
w
i
w
j
=
n

i=1
n

j=1
(c +a
i
) (c +a
j
)
a
i
+a
j
+c
w
i
w
j
=
n

i=1
n

j=1
(c +a
i
) w
i
(c +a
j
) w
j
a
i
+a
j
+c
=
n

i=1
n

j=1
v
i
v
j
a
i
+a
j
+c
(since (c +a
i
) w
i
= v
i
and (c +a
j
) w
j
= v
j
). Hence,
n

i=1
n

j=1
a
i
a
j
w
i
w
j
a
i
+a
j
+c
c
_
n

i=1
w
i
_
2
holds if and only if
n

i=1
n

j=1
v
i
v
j
a
i
+a
j
+c
0.
(9)
Also, clearly,
v
1
= v
2
= ... = v
n
= 0 holds if and only if (c +a
1
) w
1
= (c +a
2
) w
2
= ... = (c +a
n
) w
n
= 0.
(10)
By Corollary 9, the inequality
n

i=1
n

j=1
v
i
v
j
a
i
+a
j
+c
0 holds, with equality if and only
if v
1
= v
2
= ... = v
n
= 0. According to (9) and (10), this means that
n

i=1
n

j=1
a
i
a
j
w
i
w
j
a
i
+a
j
+c

c
_
n

i=1
w
i
_
2
, with equality if and only if (c +a
1
) w
1
= (c +a
2
) w
2
= ... = (c +a
n
) w
n
=
0. Thus, Corollary 10 is proven.
The problem follows from Corollary 10 (applied to c = 1 and a
i
= i).
9