Exercises from Section 1.2.3

Tord M. Johnson

February 12, 2014

1. [10] The text says that a1 + a2 + + a0 = 0. What then, is a2 + + a0?

The identities

1k0ak = 0 = a1 + 2k0ak

imply that

2k0ak = a2 + + a0 = a1.

2. [01] What does the notation 1jnaj mean, if n = 3.14?

The notation 1jnaj if n = 3.14 represents the sum of all aj such that 1 j n = 3.14, or equivalently, j {1,2,3}. That is,

1jnaj = a1 + a2 + a3.

3. [13] Without using the -notation, write out the equivalent of

0n5 1 2n + 1

and also the equivalent of

0n25 1 2n2 + 1.

Explain why the two results are different, in spite of rule (b).

For the first notation,

0n5 1 2n + 1 = 1 1 + 1 3 + 1 5 + 1 7 + 1 9 + 1 11,

while for the second notation,

0n25 1 2n2 + 1 = 1 9 + 1 3 + 1 1 + 1 3 + 1 9.

The two results are different, in spite of rule (b), as P(n) = n2 is not a permutation, since P is neither one-to-one ( 12 = 12 serves as a counterexample) nor onto (there is no integer n such that n2 = 3).

4. [10] Without using the -notation, write out the equivalent of each side of Eq. (10) as a sum of sums for the case n = 3.

For n = 3, the equivalent of the left side of Eq. (10) is

i=13 j=1ia ij = (a11) + (a21 + a22) + (a31 + a32 + a33)

and of the right side is

j=13 i=j3 = (a 11 + a21 + a31) + (a22 + a32) + (a33).

5. [HM20] Prove that rule (a) is valid for arbitrary infinite series, provided the series converge.

Proposition. The distributive law, for the products of sums ( R(i)ai)( S(j)bj) = R(i)( S(j)aibj) holds for arbitrary infinite series, provided the sums converge.

Proof. Assume that we have two infinite series ai, bj whose sums converge; that is, that

R(i)ai =(lim n R(i) 0i<n ai)+(lim n R(i) ni<0 ai) = α

and

S(j)bj =(lim n S(j) 0j<n bj)+(lim n S(j) nj<0 bj) = β

for arbitary limits α, β. We must show that

( R(i)ai)( S(j)bj) = R(i)( S(j)aibj).

But

( R(i)ai)( S(j)bj) = lim n(( R(i) 0i<n ai + R(i) ni<0 ai)( S(j) 0j<n bj + S(j) nj<0 bj)) = lim n(( R(i) 0i<n ai)( S(j) 0j<n bj + S(j) nj<0 bj) +( R(i) ni<0 ai)( S(j) 0j<n bj + S(j) nj<0 bj)) = lim n( R(i) 0i<n ai( S(j) 0j<n bj + S(j) nj<0 bj) + R(i) ni<0 ai( S(j) 0j<n bj + S(j) nj<0 bj)) = lim n( R(i) 0i<n ( S(j) 0j<n aibj + S(j) nj<0 aibj) + R(i) ni<0 ( S(j) 0j<n aibj + S(j) nj<0 aibj)) = lim n( R(i) 0i<n ( S(j)aibj) + R(i) ni<0 ( S(j)aibj)) = R(i)( S(j)aibj)

as we needed to show. □

6. [HM20] Prove that rule (d) is valid for an arbitrary infinite series, provided that any three of the four sums exist.

Proposition. Manipulating the domain, R(j)aj + S(j)aj = R(j)S(j)aj + R(j)S(j)aj is valid for arbitrary infinite series, provided that any three of the four sums exist.

Proof. Assume that R and S are arbitrary infinite series. We must show that if any three of the sums R(j)aj, S(j)aj, R(j)S(j)aj, R(j)S(j)aj exist, then we may manipulate the domain so that R(j)aj + S(j)aj = R(j)S(j)aj + R(j)S(j)aj. It is sufficient to show that the convergence of three sums is sufficient for the convergence of the fourth.

Case 1. [ R(j)aj, S(j)aj, R(j)S(j)aj converge.] We must show that R(j)S(j)aj converges. But if the sums R(j)aj, S(j)aj exist, then clearly their conjunction R(j)S(j)aj exists.

Case 2. [ R(j)aj, S(j)aj, R(j)S(j)aj converge.] We must sow that R(j)S(j)aj converges. But if the sums R(j)aj, S(j)aj exist, then clearly their disjunction R(j)S(j)aj exists.

Case 3. [ R(j)aj, R(j)S(j)aj, R(j)S(j)aj converge.] We must show that S(j)aj converges. But if the conjunction R(j)S(j)aj exists, then clearly the single sum S(j)aj exists.

Case 4. [ S(j)aj, R(j)S(j)aj, R(j)S(j)aj converge.] We must show that R(j)aj converges. But if the conjunction R(j)S(j)aj exists, then clearly the single sum R(j)aj exists.

Therefore, in all cases, we have shown that the convergence of three sums is sufficient for the convergence of the fourth. □

7. [HM23] Given that c is an integer, show that R(j)aj = R(cj)acj, even if both series are infinite.

Proposition. R(j)aj = R(cj)acj, even if both series are infinite.

Proof. Assume that R is an arbitrary relation. We must show that R(j)aj = R(cj)acj, even if both series are infinite. Since R(j) = c j is a permutation of the integers, proving the finite case, we must show the infinite case. But

R(j)aj =(lim n R(j) 0j<n aj) +(lim n R(j) nj<0 aj) =(lim n R(cj) 0cj<n acj) +(lim n R(cj) ncj<0 acj) = R(cj)acj

as we needed to show. □

8. [HM25] Find an example of infinite series in which Eq. (7) is false.

An example of an infinite series in which R(i) S(j)aij S(j) R(j)aij is given by

aij = -1if i 1 = j 1 if i + 1 = j 0 otherwise

for S(i) i 0, R(j) j 0. Then

R(i) S(j)aij = lim n 0i<n 0j<naij = lim n( 0i<1 0j<naij + 1i<n 0j<naij) = lim n(a01 + 1i<n 0j<naij) = lim n( 1 + 1i<n 0j<naij) = 1

and

S(j) R(i)aij = lim n 0j<n 0i<naij = lim n( 0j<1 0i<naij + 1j<n 0i<naij) = lim n(a10 + 1j<n 0i<naij) = lim n(1 + 1j<n 0i<naij) = 1.

9. [05] Is the derivation of Eq. (14) valid even if n = 1?

No, the derivation for Eq. (14) is not valid if n = 1, since the application of rule (d) assumes n 0. That is, if n = 1

0j1axja + 1j1axj

(depsite the fact that 0j1axj = 0 = a(1x1+1 1x )).

10. [05] Is the derivation of Eq. (14) valid even if n = 2?

No, the derivation for Eq. (14) is not valid if n = 2, since the application of rule (d) assumes n 0. That is, if n = 2

0j2axja + 1j2axj.

11. [03] What should the right-hand side of Eq. (14) be if x = 1?

In the case x = 1

0jnaxj = 0jna(1)j = 0jna = (n + 1)a.

12. [10] What is 1 + 1 7 + 1 49 + 1 343 + + (1 7)n?

The sum 1 + 1 7 + 1 49 + 1 343 + + (1 7)n is simply a geometric progression whose closed form is given by Eq. (14) with a = 1 and x = 1 7. This yields

0jn(1) 1 7j = (1) 1 1 7 n+1 1 1 7 = 1 1 7 n+1 67 = 7 6 1 17n+1 .

13. [10] Using Eq. (15) and assuming that m n, evaluate j=mnj.

Since

mjnj = 0jnj 0jm1j

we can use Eq. (15) with a = 0 and b = 1 to evaluate each term on the right-hand side as

mjnj = 0jnj 0jm1j = 1 2n(n + 1) 1 2(m 1)m = 1 2(n(n + 1) m(m 1)).

14. [11] Using the result of the previous exercise, evaluate j=mn k=rsjk.

Since mjnj = 1 2(n(n + 1) m(m 1)), we can evaluate the sum as

mjn rksjk = mjnj rksk = 1 2(n(n + 1) m(m 1)) 1 2(s(s + 1) r(r 1)) = 1 4 (n(n + 1) m(m 1))(s(s + 1) r(r 1)).

15. [M22] Compute the sum 1 × 2 + 2 × 22 + 3 × 23 + + n × 2n for small values of n. Do you see the pattern developing in these numbers? If not, discover it by manipulations similar to those leading up to Eq. (14).

We can manipulate the sum as

0knk2k = 0 + 1knk2k = 1knk2k = 2 1knk2k1 = 2 0kn1(k + 1)2k = 2 0kn1k2k + 0kn12k = 2 0knk2k n2n + 0kn12k = 2 0knk2k n2n (1 2n) = 2 0knk2k n2n+1 (2 2n+1).

Comparing the first with the last yields

0knk2k = n2n+1 + 2 2n+1 = 2(n2n 2n + 1).

16. [M22] Prove that

j=0njxj = nxn+2 (n + 1)xn+1 + x (x 1)2 ,

if x1, without using mathematical induction.

Proposition. 0jnjxj = nxn+2(n+1)xn+1+x (x1)2 if x1.

Proof. Assume that x1, n 0. Then

0jnjxj = 0 + 1jnjxj = 1jnjxj = x 1jnjxj1 = x 0jn1(j + 1)xj = x 0jn1jxj + 0jn1xj = x 0jnjxj nxn + 0jn1xj = x 0jnjxj nxn + 1 xn 1 x = x 0jnjxj nxn+1 + x1 xn 1 x .

Comparing the first relation with the last, we have

(1 x) 0jnjxj = nxn+1 + x1 xn 1 x ;

hence we obtain the formula

0jnjxj = (1 x)nxn+1 (1 x)2 + x 1 xn (1 x)2 = nxn+1 + nxn+2 + x xn+1 (1 x)2 = nxn+2 (n + 1)xn+1 + x (1 x)2

as we needed to show. □

17. [M00] Let S be a set of integers. What is jS1?

jS1 is the cardinality—the number of elements—of S; that is,

jS1 = |S|.

18. [M20] Show how to interchange the order of summation as in Eq. (9) given that R(i) is the relation “n is a multiple of i” and S(i,j) is the relation “1 j < i.”

In Eq. (9), we have

R(i) S(i,j)aij = S(j) R(i,j)aij

for S(j) (i)(R(i) S(i,j)) and R(i,j) R(i) S(i,j).

In the case that R(i) (k)(ki = n) and S(i,j) 1 j < i, we find

S(j) (i)(R(i) S(i,j)) (i)((k)(ki = n) 1 j < i) 1 j < n for i nk 1

and

R(i,j) R(i) S(i,j) (k)(ki = n) 1 j < i (k)(ki = n) j < i;

that is, S(j) the relation “1 j < n” and R(i,j) the relation “n is a multiple of i and i > j.”

19. [20] What is j=mn(aj aj1)?

We have that

mjn(aj aj1) = mjnaj mjnaj1 = mjnaj m1jn1aj = mjnaj am1 mjnaj + an = an am1

assuming m n.

20. [25] Dr. I. J. Matrix has observed a remarkable sequence of formulas:

9 × 1 + 2 = 11,9 × 12 + 3 = 111,9 × 123 + 4 = 1111,9 × 1234 + 5 = 11111.
a.
Write the good doctor’s great discovery in terms of the -notation.
b.
Your answer to part (a) undoubtedly involves the number 10 as a base of the decimal system; generalize this formula so that you get a formula that will perhaps work in any base b.
c.
Prove your formula from part (b) by using formulas derived in the text or in exercise 16 above.

The remarkable sequence of formulas observed by Dr. I. J. Matrix are analyzed below.

a.
The good doctor’s great discovery in terms of the -notation is
(10 1) 0kn(n k)10k + n + 1 = 0kn10k.
b.
The above formula may be generalized for perhaps any base b as
(b 1) 0kn(n k)bk + n + 1 = 0knbk.
c.
We may prove the above formula.

Proposition. (b 1) 0kn(n k)bk + n + 1 = 0knbk.

Proof. Assume an arbitrary base b 2, and n 0. We must show that

(b 1) 0kn(n k)bk + n + 1 = 0knbk.

But

(b 1) 0kn(n k)bk + n + 1 = (b 1) n 0knbk 0knkbk + n + 1 = (b 1) n 1 bn+1 1 b 0knkbk + n + 1 by Eq. (14) = (b 1) n 1 bn+1 1 b nbn+2 (n + 1)bn+1 + b (b 1)2 + n + 1by exercise 16 = n(bn+1 1) + b(bn(nb + n + 1) 1) b 1 + n + 1 = bn+1 + n b(n + 1) b 1 + n + 1 = 1 bn+1 1 b = 0knbk by Eq. (14)

as we needed to show. □

21. [M25] Derive rule (d) from (8) and (17).

We may derive rule (d) for manipulating the domain

R(j)aj + S(j)aj = R(j)S(j)aj + R(j)S(j)aj

from Eq. (8)

R(i)(bi + ci) = R(i)bi + R(i)ci

and Eq. (17)

R(j)aj = jaj[R(j)].

Given that [p] + [q] = [p q] + [p q], as evidenced by the following table

[p][q][p] + [q][p q][p q][p q] + [p q]






000000
0 1 1 1 0 1
101101
1 1 2 1 1 2

we have

R(j)aj + S(j)aj = jaj[R(j)] + jaj[S(j)] by Eq. (17) = j(aj[R(j)] + aj[S(j)]) by Eq. (8) = jaj([R(j)] + [S(j)]) = jaj([R(j) S(j)] + [R(j) S(j)]) = jaj[R(j) S(j)] + jaj[R(j) S(j)] by Eq. (8) = R(j)S(j)aj + R(j)S(j)aj. by Eq. (17)

22. [20] State the appropriate analogs of Eqs. (5), (7), (8), and (11) for products instead of sums.

We have the following analogs for products: change of variable:

R(i)ai = R(j) = R(p(j))ap(j);

interchanging order of production:

R(i) S(j)aij = S(j) R(i)aij;

a special case of the above:

R(i)(bici) = R(i)bi R(i)ci ;

and manipulating the domain:

R(j)aj S(j)aj = R(j)S(j)aj R(j)S(j)aj .

23. [10] Explain why it is a good idea to define R(j)aj and R(j)aj as zero and one, respectively, when no integers satisify R(j).

It is a good idea to define jaj = 0 and jaj = 1 as they are the identity elements for the operations of addition and multiplication, respectively. This way, jaj + R(j)aj = R(j)aj and jaj R(j)aj = R(j)aj.

24. [20] Suppose that R(j) is true for only finitely many j. By induction on the number of integers satisfying R(j), prove that log b R(j)aj = R(j)(log baj), assuming that all aj > 0.

Proposition. log b R(j)aj = R(j)(log baj) for all aj > 0.

Proof. Suppose aj > 0 for 0 j n. We must show that log b 0jnaj = 0jn(log baj).

If n = 0, clearly log b 0j0aj = log ba0 = 0j0(log baj). Then, assuming

log b 0jkaj = 0jk(log baj),

we must show that

log b 0jk+1aj = 0jk+1(log baj).

But

log b 0jk+1aj = log b ak+1 0jkaj = log b 0jkaj + log bak+1 = 0jk(log baj) + log bak+1 = 0jk+1(log baj)

as we needed to show. □

25. [15] Consider the following derivation; is anything amiss?

( i=1na i)( j=1n 1 aj) = 1in 1jn ai aj = 1in 1inai ai = i=1n1 = n.

Yes. For one,

1in 1jn ai aj 1in 1inai ai;

and for another,

1in 1inai ai i=1n1

(in fact, 1in 1inai ai = i=1nn = n2).

26. [25] Show that i=0n j=0iaiaj may be expressed in terms of i=0nai by manipulating the -notation as stated in exercise 22.

Proposition. 0in 0jiaiaj = 0inai n+2.

Proof. We must show that 0in 0jiaiaj = 0inai n+2.

First note that

0in 0jiaiaj = 0in ijnaiaj;

and then that

0in 0jiaiaj 2 = 0in 0jiaiaj 0in 0jiaiaj = 0in 0jiaiaj 0jiaiaj = 0in 0jiaiaj ijnaiaj = 0in 0jnaiaj aiai = 0in 0jnaiaj 0inai2 = 0in ain+1 0jnaj 0inai 2 = 0inai n+1 0in 0jnaj 0inai 2 = 0inai n+1 0jnaj n+1 0inai 2 = 0inai 2n+4.

Therefore,

0in 0jiaiaj = 0inai n+2

as we needed to show. □

27. [M20] Generalize the result of exercise 1.2.1-9 by proving that

j=1n(1 a j) 1 j=1na j,

assuming that 0 < aj < 1.

Proposition. 1jn(1 aj) 1 1jnaj if 0 < aj < 1.

Proof. Let 0 < aj < 1 for 1 j n, n 0. We must show that

1jn(1 aj) 1 1jnaj.

If n = 0, then clearly 1jn(1 aj) = 1 1 = 1 1jnaj. Then, assuming

1jk(1 aj) 1 1jkaj,

we must show that

1jk+1(1 aj) 1 1jk+1aj.

But since ak+1 0jkak 1,

1jk+1(1 aj) = 1jk(1 aj)(1 ak+1) 1 1jkaj (1 ak+1) = 1 1jkaj ak+1 ak+1 0jkaj = 1 0jkaj ak+1 + ak+1 0jkak 1 0jkaj + ak+1 = 1 1jk+1aj

as we needed to show. □

28. [M22] Find a simple formula for j=2n(1 1j2).

We find that

2jn(1 1 j2) = 2jnj2 1 j2 = 2jn 1 j2 2jn(j 1)(j + 1) = 2jn1 j2 1jn1j 3jn+1j = 1 n!2(n 1)!(n + 1)! 2 = n + 1 2n .

29. [M30] (a) Express i=0n j=0i k=0jaiajak in terms of the multiple-sum notation explained at the end of the section. (b) Express the same sum in terms of i=0nai, i=0nai2, and i=0nai3 [see Eq. (13)].

a.
We can express the sum in terms of the multiple-sum notation as
0in 0ji 0kjaiajak = 0kjinaiajak.
b.
From Eq. (13) for arbitrary i > 0 we have
0ji1 0kjajak = 1 2 0ji1aj 2 + 0ji1aj2

or equivalently

0ji1aj 2 = 2 0ji1 0kjajak 0ji1aj2.

Also for arbitrary i > 0 we have that

0jiaj 3 = 0ji1aj + ai 3 = 0ji1aj 3 + 3a i 0ji1aj 2 + 3a i2 0ji1aj + ai3 = 0ji1aj 3 + 3a i 2 0ji1 0kjajak 0ji1aj2 + 3a i2 0ji1aj + ai3 = 0ji1aj 3 + 6 0ji1 0kjaiajak 3 0ji1aiaj2 + 3 0ji1ai2a j + ai3 = 0ji1aj 3 + 6 0ji1 0kjaiajak 3 0jiaiaj2 + 3 0jiai2a j + ai3 = 0ji1aj 3 + 6 0ji 0kjaiajak 6 0jiai2a j 3 0jiaiaj2 + 3 0jiai2a j + ai3 = 0ji1aj 3 + 6 0ji 0kjaiajak 3 0jiaiaj2 3 0jiai2a j + ai3 = 0ji1aj 3 + 6 0ji 0kjaiajak 3 0jiaiaj(ai + aj) + ai3

and in the trivial case for i = 0 that

0jiaj 3 = a 03 = (0)3 + 6a 03 3(2)a 03 + a 03

so that

6 0ji 0kjaiajak = 0jiaj 3 0ji1aj 3 + 3 0jiaiaj(ai + aj) ai3

and so that for 0 i n we have 6 0in 0ji 0kjaiajak equivalent to

0in 0jiaj 3 0ji1aj 3 + 3 0in 0jiaiaj(ai + aj) 0inai3.

We may also prove by induction that

0in 0jiaj 3 0ji1aj 3 = 0inai 3.

If n = 0, clearly a03 (0)3 = a03. Then, assuming

0ik 0jiaj 3 0ji1aj 3 = 0ikai 3

we must show that

0ik+1 0jiaj 3 0ji1aj 3 = 0ik+1ai 3.

But

0ik+1 0jiaj 3 0ji1aj 3 = 0ik 0jiaj 3 0ji1aj 3 + 0jk+1aj 3 0jkaj 3 = 0ikai 3 + 0jk+1aj 3 0jkaj 3 = 0jk+1aj 3

And so finally, we have that

6 0in 0ji 0kjaiajak = 0inai 3 + 3 0in 0jiaiaj(ai + aj) 0inai3

and

6 0in 0ji 0kjaiajak = 0inai 3 + 3 0in 0jiaiaj(ai + aj) 0inai3 = 0inai 3 + 0in 0jiaiaj(ai + aj) 0inai3 + 2 0in 0jiaiaj(ai + aj) = 0inai 3 + 0in 1 2 0jiaiaj(ai + aj) + 1 2 ijnaiaj(ai + aj) 0inai3 + 2 0in 0jiaiaj(ai + aj) = 0inai 3 + 0in 1 2 0jnaiaj2 + 1 2 0jnai2a j + ai3 0inai3 + 2 0in 0jiaiaj(ai + aj) = 0inai 3 + 0inai 0inai2 + 0inai3 0inai3 + 2 0in 0jiaiaj(ai + aj) = 0inai 3 + 0inai 0inai2 + 2 0in 0jiaiaj(ai + aj) = 0inai 3 + 0inai 0inai2 + 0in 0jiaiaj(ai + aj) + ijnaiaj(ai + aj) = 0inai 3 + 0inai 0inai2 + 0in 0jnaiaj(ai + aj) + 2ai3 = 0inai 3 + 0inai 0inai2 + 0in 0jnaiaj2 + 0in 0jnai2a j + 2 0inai3 = 0inai 3 + 0inai 0inai2 + 2 0inai 0inai2 + 2 0inai3 = 0inai 3 + 3 0inai 0inai2 + 2 0inai3

Therefore,

0in 0ji 0kjaiajak = 1 6 0inai 3 + 1 2 0inai 0inai2 + 1 3 0inai3.

30. [M23] (J. Binet, 1812.) Without using induction, prove the identity

( j=1na jxj)( j=1nb jyj) =( j=1na jyj)( j=1nb jxj)+ 1jkn(ajbkakbj)(xjykxkyj).

[An important special case arises when w1,,wn,z1,,zn are aribtrary complex numbers and we set aj = wj, bj = z̄j, xj = w̄j, yj = zj:

( j=1n|w j|2)( j=1n|z j|2) =| j=1nw jzj|2 + 1jkn|wjz̄k wkz̄j|2.

The terms |wjz̄j|2 are nonnegative, so the famous Cauchy-Schwarz inequality

( j=1n|w j|2)( j=1n|z j|2) | j=1nw jzj|2

is a consequence of Binet’s formula.]

Proposition. 1jnajxj 1jnbjyj = 1jnajyj 1jnbjxj + 1jn j<kn(ajbk akbj)(xjyk xkyj).

Proof. We need to show that

1jnajxj 1jnbjyj = 1jnajyj 1jnbjxj+ 1jn j<kn(ajbkakbj)(xjykxkyj).

But

1jn j<kn(ajbk akbj)(xjyk xkyj) = 1jn j<knajbk(xjyk xkyj) 1jn j<knakbj(xjyk xkyj) = 1jn j<knajbk(xjyk xkyj) + 1jn j<knakbj(xkyj xjyk) = 1jn j<knajbk(xjyk xkyj) + 1kn k<jnajbk(xjyk xkyj) = 1jn j<knajbk(xjyk xkyj) + 1jn 1k<jajbk(xjyk xkyj) = 1jn 1k<jajbk(xjyk xkyj) + 0 + 1jn j<knajbk(xjyk xkyj) = 1jn 1k<jajbk(xjyk yjxk) + 1jnajbj(xjyj yjxj) + 1jn j<knajbk(xjyk yjxk) = 1jn 1knajbk(xjyk yjxk) = 1jn 1knajxjbkyk 1jn 1knajyjbkxk = 1jnajxj 1jnbjyj 1jnajyj 1jnbjxj

as we needed to show. □

31. [M20] Use Binet’s formula to express the sum 1jkn(uj uk)(vj vk) in terms of j=1nujvj, j=1nuj, and j=1nvj.

We want to find an expression for

1jn j<kn(uj uk)(vj vk)

in terms of 1jnujvj, 1jnuj, and 1jnvj.

From Binet’s formula we have that

1jn j<kn(ajbkakbj)(xjykxkyj) = 1jnajxj 1jnbjyj 1jnajyj 1jnbjxj .

If we let aj = uj, xj = vj, and bj = yj = 1, we find

1jn j<kn(uj uk)(vj vk) = 1jnujvj 1jn1 1jnuj 1jnvj = n 1jnujvj 1jnuj 1jnvj .

______________________________________________________________________________________________________________________________

[See Soobschch. Mat. Obschch. Khar’kovskom Univ. 4, 2 (1882), 93–98.]

32. [M20] Prove that

j=1n i=1ma ij = 1i1,,inmai11ainn.

Proposition. 1jn 1ijmaijj = 1i1,,inmai11ainn.

Proof. We need to show that

1jn 1ijmaijj = 1i1,,inmai11ainn.

If n = 1, clearly 1i1mai11 = 1i1mai11. Then, assuming that

1jk 1ijmaijj = 1i1,,ikmai11aikk

we must show that

1jk+1 1ijmaijj = 1i1,,ik+1mai11aik+1(k+1).

But

1jk+1 1ijmaijj = 1jk 1ijmaijj 1ik+1maik+1(k+1) = 1i1,,ikmai11aikk 1ik+1maik+1(k+1) = 1i1,,ik+1mai11aik+1(k+1)

as we needed to show. □

33. [M30] One evening Dr. Matrix discovered some formulas that might even be classed as more remarkable than those of exercise 20:

1 (a b)(a c) + 1 (b a)(b c) + 1 (c a)(c b) = 0,
a (a b)(a c) + b (b a)(b c) + c (c a)(c b) = 0,
a2 (a b)(a c) + b2 (b a)(b c) + c2 (c a)(c b) = 1,
a3 (a b)(a c) + b3 (b a)(b c) + c3 (c a)(c b) = a + b + c.

Prove that these formulas are a special case of a general law; let x1,x2,,xn be distinct numbers, and show that

j=1n(x jr 1kn kj (xjxk)) = 0, if 0 r < n 1, 1, if r = n 1; j=1nxj,if r = n.

Proposition. 1in xir 1jn ji (xixj) = 0 if 0 r < n 1 1 if r = n 1 1inxiif r = n if x1,x2,,xn distinct.

Proof. For an arbitrary series of distinct numbers xi, 1 i n, and for an arbitrary ι, 1 ι n, let P(xι) = xιr for 0 r n, and Q(xι) = 1in iι (xιxi). By the fundamental theorem of algebra and the method of partial fractions, since r = deg P deg Q + 1 = n, we have that

P(xι) Q(xι) = xιr 1in iι (xι xi) = D(xι)+ 1in iι ci xι xi

for constants ci, where D(xι) is the polynomial divisor with remainder R(xι) such that

P(xι) = D(xι)Q(xι) + R(xι)deg R < deg Q

By polynomial division, we have

D(xι) = 0 if 0 r = deg P < deg Q = n 1 1 if r = deg P = deg Q = n 1 1inxiif r = deg P = deg Q + 1 = n

Also, for an arbitrary κ, 1 κ n, we have

xιr 1in iι (xι xi) = D(xι) + 1in iι ci xι xi xιr (xι xκ) 1in iι iκ (xι xi) = D(xι) + 1in iι iκ ci xι xi + cκ xι xκ xιr(xι xκ) (xι xκ) 1in iι iκ (xι xi) = (xι xκ)D(xι) + (xι xκ) 1in iι iκ ci xι xi + cκ(xι xκ) xι xκ xιr 1in iι iκ (xι xi) = (xι xκ)D(xι) + (xι xκ) 1in iι iκ ci xι xi + cκ cκ = xιr 1in iι iκ (xι xi) + (xκ xι)D(xι) + (xκ xι) 1in iι iκ ci xι xi

Letting ι = κ, we find

cκ = xκr 1in iκ (xκ xi) + (xκ xκ)D(xκ) + (xκ xκ) 1in iκ ci xκ xi = xκr 1in iκ (xκ xi)

And so

xιr 1in iι (xι xi) = D(xι) + 1in iι ci xι xi = D(xι) + 1in iι xir (xι xi) 1jn ji (xi xj) = D(xι) 1in iι xir (xi xι) 1jn ji (xi xj)

or equivalently

D(xι) = xιr 1in iι (xι xi)+ 1in iι xir (xi xι) 1jn ji (xi xj)

letting ι = n yields

D(xn) = xnr 1in in (xn xi) + 1in in xir (xi xn) 1jn ji (xi xj) = 1in1 xir (xi xn) 1jn1 ji (xi xj) + xnr 1in1 (xn xi) = 1in1 1jn1 ji 1kn1 kj (xj xk) xir xixn 1in1 1jn1 ji (xi xj) + xnr 1in1 (xn xi) = 1in1 1 xixn 1jn1 ji 1kn1 kj (xj xk)xir 1in1 1jn1 ji (xi xj) + xnr 1in1 (xn xi) = 1in1 1jn1 ji (xj xn) 1jn1 ji 1kn1 kj (xj xk)xir 1in1(xi xn) 1in1 1jn1 ji (xi xj) + xnr 1in1 (xn xi) = 1in1 1jn1 (xn xj) 1jn1 ji (xj xn) 1jn1 ji 1kn1 kj (xj xk)xir 1jn1 (xn xj) 1in1(xi xn) 1in1 1jn1 ji (xi xj) + 1in1 (xi xn) 1in1 1jn1 ji (xi xj)xnr 1jn1 (xn xj) 1in1(xi xn) 1in1 1jn1 ji (xi xj) = U V

where

U = 1in1 1jn1 (xn xj) 1jn1 ji (xj xn) 1jn1 ji 1kn1 kj (xj xk)xir + 1in1 (xi xn) 1in1 1jn1 ji (xi xj)xnr = 1in1 1jn1 (xn xj) 1jn1 ji (xj xn) 1jn1 ji 1kn1 kj (xj xk)xir + 1in1 1jn1 ji (xi xj)(xi xn)xnr = 1in1 1jn1 (xn xj) 1jn1 ji (xj xn) 1jn1 ji 1kn1 kj (xj xk)xir + 1in1 1jn ji (xi xj)xnr = 1in1 1jn1 ji 1kn1 kj (xj xk)(xj xn) 1kn kn (xn xk)xir + 1jn jn 1kn kj (xj xk)xnr = 1in1 1jn1 ji 1kn kj (xj xk) 1kn kn (xn xk)xir + 1jn jn 1kn kj (xj xk)xnr = 1in1 1jn ji 1kn kj (xj xk)xir + 1jn jn 1kn kj (xj xk)xnr = 1in 1jn ji 1kn kj (xj xk)xir

and where

V = 1jn1 (xn xj) 1in1(xi xn) 1in1 1jn1 ji (xi xj) = 1in1 1jn1 ji (xi xj)(xi xn) 1jn1 (xn xj) = 1in1 1jn ji (xi xj) 1jn1 (xn xj) = 1in 1jn ji (xi xj)

so that

D(xn) = U V = 1in 1jn ji 1kn kj (xj xk)xir 1in 1jn ji (xi xj) = 1in xir 1jn ji (xi xj).

That is,

1in xir 1jn ji (xi xj) = 0 if 0 r < n 1 1 if r = n 1 1inxiif r = n

as we needed to show. □

______________________________________________________________________________________________________________________________

Notes: Dr. Matrix was anticipated in this discovery by L. Euler, who wrote to Christian Golbach about it on 9 November 1762. See Euler’s Institutionum Calculi Integralis 2 (1769), §1169; and E. Waring, Phil. Trans. 69 (1779), 64–67 [J. J. Sylvester, Quart. J. Math. 1 (1857), 141–152.]

34. [M25] Prove that

k=1n 1rnrm(x + k r) 1rnrk(k r) = 1,

provided that 1 m n and x is arbitrary. For example, if n = 4 and m = 2, then

x(x 2)(x 3) (1)(2)(3) + (x + 1)(x 1)(x 2) (1)(1)(2) + (x + 2)x(x 1) (2)(1)(1) + (x + 3)(x + 1)x (3)(2)(1) = 1.

Proposition. 1in 1jnjk(x+ij) 1jnji(ij) = 1 provided that 1 k n and x is arbitrary.

Proof. We may prove

1in 1jnjk(x + i j) 1jnji(i j) = 1

provided that 1 k n and x is arbitrary, but first we shall first prove the more general result

1in 1jn,jk(yi zj) 1jn,ji(yi yj) = 1.

Let P(y) be the polynomial representation of 1jn,jk(y zj) where

P(y) = 1jn jk (yzj) = 0jn1cjyj

for arbitrary coefficients c0,c1,,cn1. Note that since P(y) = yn1 + , cn1 = 1. Then, from exercise 33, we find that

1in 1jn,jk(yi zj) 1jn,ji(yi yj) = 1in P(yi) 1jn,ji(yi yj) = 1in 0jn1cjyij 1jn,ji(yi yj) = 1jn 0in1ciyji 1kn,kj(yj yk) = 0in1ci 1jn yji 1kn,kj(yj yk) = 0i<n1ci 1jn yji 1kn,kj(yj yk) + cn1 1jn yjn1 1kn,kj(yj yk) = 0i<n1ci 0 + cn1 1 = cn1 = 1.

Letting yi = i and zi = i + x for 1 i n, 1 k n, and x arbitrary, we have that

1in 1jnjk(x + i j) 1jnji(i j) = 1

as we needed to show. □

35. [HM20] The notation sup R(j)aj is used to denote the least upper bound of the elements aj, in a manner analogous to the - and -notations. (When R(j) is satisfied for only finitely many j, the notation max R(j)aj is often used to denote the same quantity.) Show how rules (a), (b), (c), and (d) can be adapted for manipulation of this notation. In particular discuss the following analog of rule (a):

(sup R(j)ai) + (sup S(j)bj) = sup R(i)(sup S(j)(ai + bj)),

and give a suitable definition for the notation when R(j) is satisfied for no j.

We have the following analogs for the least upper bound: an additive law:

(sup R(i)ai) + (sup S(j)bj) = sup R(i)(sup S(j)(ai + bj))

as well as a multiplicative law:

(sup R(i)ai)(sup S(j)bj) = sup R(i)(sup S(j)(aibj))

provided ai and aj are nonnegative; change of variable:

sup R(i)ai = sup R(j) = sup R(p(j))ap(j);

interchanging order of bound:

sup R(i) sup S(j)aij = sup S(j) sup R(i)aij;

and manipulating the domain:

sup (sup R(j)aj,sup S(j)aj) = sup R(j)S(j)aj.

A suitable definition for the notation when R(j) is satisfied for no j would be

sup j =

since acts as the identity element for the least upper bound, in that sup (sup jaj,sup R(j)aj) = sup R(j)aj.

36. [M23] Show that the determinant of the combinatorial matrix is xn1(x + ny).

Proposition. det [y + δijx]n = xn1(ny + x).

Proof. For

An = [aij]n = [y+δijx]n = y + x y y y y + x y y y y + x n,

we must show that

det [aij]n = xn1(ny + x).

Let

aij = ai1 if j = 1 aij ai1if 2 j n

and

aij = a1j + 2knakjif i = 1 aij if 2 i n

so that det [aij]n = det [aij ]n where

aij = ny + xif i = 1j = 1 0 if i = 12 j n y if 2 i nj = 1 δijx if 2 i n2 j n

since:

a11 = a 11 = y + x i = 1,j = 1 a1j = x i = 1,2 j n ai1 = a i1 = y 2 i n,j = 1 aij = δ ijx 2 i n,2 j n a11 = a 11 + 2knak1 = y + x + 2kny = ny + x i = 1,j = 1 a1j = a 1j + 2knakj = x + 2knδkjx = 0 i = 1,2 j n ai1 = a i1 = y 2 i n,j = 1 aij = a ij = δ ijx 2 i n,2 j n

Then, since [aij]n is a triangular matrix (aij = 0 whenever i < j), we have that

det [aij]n = det [aij ] n = 1knakk = a11 2knakk = (ny + x) 2knδkkx = (ny + x) 2knx = (ny + x)xn1 = xn1(ny + x)

as we needed to show. □

37. [M24] Show that the determinant of Vandermonde’s matrix is

1jnxj 1i<jn(xj xi).

Proposition. det [xji]n = 1inxi 1j<in(xi xj).

Proof. For

An = [aij]n = [xji] n = x1 x2 xn x12x22xn2 x 1nx 2nx nn n,

we must show that

det [aij]n = 1inxi 1j<in(xi xj).

Let

aij = ai1 if j = 1 aij ai1if 2 j n

and

aij = a11 if i = 1,j = 1 a1j if i = 1,2 j n ai1 x1a(i1)1if 2 i n,j = 1, aij x1a(i1)jif 2 i n,2 j n,

so that det [aij] = det [aij ] where

aij = x1 if i = 1j = 1 xj x1 if i = 12 j n 0 if 2 i nj = 1 xji1(xj x1)if 2 i n2 j n

since:

a11 = a 11 = x1 i = 1,j = 1 a1j = a 1j a11 = xj x1 i = 1,2 j n ai1 = a i1 = x1i 2 i n,j = 1 aij = a ij ai1 = xji x 1i 2 i n,2 j n a11 = a 11 = a 11 = x1 i = 1,j = 1 a1j = a 1j = a 1j a11 = xj x1 i = 1,2 j n ai1 = a i1 x 1a(i1)1 = a i1 x1a(i1)1) = 0 2 i n,j = 1 aij = a ij x 1a(i1)j = a ij ai1 x1(a(i1)j a(i1)1) = xji1(x j x1)2 i n,2 j n

Let

qij = xj+1 x1if i = j,1 i,j n 1 0 otherwise

so that minor ([a ]n,1,1) = ([qij]n1 minor ([a]n,1,1)T)T and det [qij]n1 = 1kn1(xk+1 x1). Then

det [aij]n = det [aij ] n = 1inai1 cofactor (a i1 ) = a11 cofactor (a 11 ) + 2inai1 cofactor (a i1 ) = x1(1)1+1 det minor ([a ] n,1,1) + 0 = x1 det minor ([a ]n,1,1) = x1 det (([qij]n1 minor ([a]n,1,1)T)T) = x1 det ([qij]n1 minor ([a]n,1,1)T) = x1 det [qij]n1 det (minor ([a]n,1,1)T) = x1 det [qij]n1 det minor ([a]n,1,1) = x1 1kn1(xk+1 x1)det minor ([a]n,1,1).

We shall finally use this recursive identity to give a proof by mathematical induction on n.

If n = 1, then clearly

det [aij]1 = det [a11]1 = x1 = 1inxi 1j<i1(xi xj).

Then, assuming that

det [aij]k = 1ikxi 1j<ik(xi xj)

or equivalently that

det minor ([a]k+1,1,1) = 2ik+1xi 2j<ik+1(xi xj)

we must show that

det [aij]k+1 = 1ik+1xi 1j<ik+1(xi xj).

But

det [aij]k+1 = x1 1ik(xi+1 x1)det [a(i+1)(j+1)]k = x1 1ik(xi+1 x1) 2ik+1xi 2j<ik+1(xi xj) = x1 2ik+1(xi x1) 2ik+1xi 2j<ik+1(xi xj) = 1i1xi 1j<ik+1(xi xj) 2ik+1xi 2j<ik+1(xi xj) = 1ik+1xi 1j<ik+1(xi xj)

as we needed to show. □

38. [M25] Show that the determinant of Cauchy’s matrix is

1i<jn(xj xi)(yj yi) 1i,jn(xi + yi).

Proposition. det [1(xi + yj)]n = 1i<jn(xj xi)(yj yi) 1i,jn(xi + yj).

Proof. For

An = [aij]n = [1(xi+yi)]n = 1(x1 + y1)1(x1 + y2)1(x1 + yn) 1(x2 + y1)1(x2 + y2)1(x2 + yn) 1(xn + y1)1(xn + y2)1(xn + yn) n,

we must show that

det [aij]n = 1i<jn(xj xi)(yj yi) 1i,jn(xi + yj).

Let

aij = ai1 if j = 1 aij ai1if 2 j n

so that det [aij] = det [aij] where

aij = 1(x1 + y1) if i = 1j = 1 ((y1 yj)(x1 + y1))(1(x1 + yj))if i = 12 j n 1(x1 + y1) if 2 i nj = 1 ((y1 yj)(xi + y1))(1(xi + yj)) if 2 i n2 j n

since:

a11 = a 11 = 1(x1 + y1) i = 1,j = 1 a1j = a 1j a11 = ((y1 yj)(x1 + y1))(1(x1 + yj)) i = 1,2 j n ai1 = a i1 = 1(xi + y1) 2 i n,j = 1 aij = a ij ai1 = ((y1 yj)(xi + y1))(1(xi + yj)) 2 i n,2 j n

Let

bij = 1 if j = 1 1(xi + yj)otherwise,

pij = 1(xi + y1)if i = j,1 i,j n 0 otherwise, and qij = 1 if i = j = 1 y1 yjif i = j,2 i,j n 0 otherwise

so that

[aij] n = ([qij]n([pij]n[bij]n)T)T

and:

det [pij]n = 1kn 1 xk + y1 det [qij]n = 2kny1 yk

Also let

bij = b1j if i = 1 bij b1jif 2 i n

so that det [bij] = det [bij] where

bij = 1 if i = 1j = 1 1(xi + yj) if i = 12 j n 0 if 2 i nj = 1 ((x1 xi)(x1 + yj))(1(xi + yj))if 2 i n2 j n

since:

b11 = b 11 = 1 i = 1,j = 1 b1j = b 1j = 1(xi + yj) i = 1,2 j n bi1 = b i1 b11 = 0 2 i n,j = 1 bij = b ij b1j = ((x1 xi)(x1 + yj))(1(xi + yj)) 2 i n,2 j n

Also let

rij = 1 if i = j = 1 x1 xiif i = j,2 i,j n 0 otherwise

and

sij = 1 if i = j = 1 1(x1 + yj)if i = j,2 i,j n 0 otherwise

so that

minor ([b] n,1,1) = ([sij]n1([rij]n1 minor ([a]n,1,1))T)T

and:

det [rij]n1 = 2knx1 xk det [sij]n1 = 2kn 1 x1 + yj

Then

det [aij] = det [aij ] n = det (([qij]n([pij]n[bij]n)T)T) = det ([qij]n)det ([pij]n[bij]n) = det ([qij]n)det ([pij]n)det ([bij]n) = 2kny1 yk det ([pij]n)det ([bij]n) = 2kny1 yk 1kn 1 xk + y1 det ([bij]n) = 2kny1 yk 1kn 1 xk + y1 det ([bij] n) = 2kny1 yk 1kn 1 xk + y1 b11cofactor (b 11) + 2inbi1cofactor (b i1) = 2kny1 yk 1kn 1 xk + y1 b11cofactor (b 11) + 0 = 2kny1 yk 1kn 1 xk + y1 det (minor ([b] n,1,1)) + 0 = 2kny1 yk 1kn 1 xk + y1 det (([sij]n1([rij]n1 minor ([a]n,1,1))T)T) = 2kny1 yk 1kn 1 xk + y1 det ([sij]n1)det ([rij]n1 minor ([a]n,1,1)) = 2kny1 yk 1kn 1 xk + y1 det ([sij]n1)det ([rij]n1)det (minor ([a]n,1,1)) = 2kny1 yk 1kn 1 xk + y1 2kn 1 x1 + yj det ([rij]n1)det (minor ([a]n,1,1)) = 2kny1 yk 1kn 1 xk + y1 2kn 1 x1 + yj 2knx1 xk det (minor ([a]n,1,1)) = 2in(xi x1)(yi y1) 1i,jn(xi + y1)(x1 + yj)det (minor ([a]n,1,1)).

We shall finally use this recursive identity to give a proof by mathematical induction on n.

If n = 1, then clearly

det [aij]1 = det [a11]1 = 1(x1 + y1) = 1i<j1(xj xi)(yj yi) 1i,j1(xi + yj).

Then, assuming that

det [aij]k = 1i<jk(xj xi)(yj yi) 1i,jk(xi + yj)

or equivalently that

det minor ([a]k+1,1,1) = 2i<jk+1(xj xi)(yj yi) 2i,jk+1(xi + yj)

we must show that

det [aij]k+1 = 1i<jk+1(xj xi)(yj yi) 1i,jk+1(xi + yj).

But

det [aij]k+1 = 2ik+1(xi x1)(yi y1) 1i,jk+1(xi + y1)(x1 + yj)det minor ([a]k+1,1,1) = 2ik+1(xi x1)(yi y1) 1i,jk+1(xi + y1)(x1 + yj) 2i<jk+1(xj xi)(yj yi) 2i,jk+1(xi + yj) = 2jk+1(xj x1)(yj y1) 2i<jk+1(xj xi)(yj yi) 1i,jk+1(xi + y1)(x1 + yj) 2i,jk+1(xi + yj) = 1i<jk+1(xj xi)(yj yi) 2i<jk+1(xj xi)(yj yi) 1i,jk+1(xi + y1)(x1 + yj) 2i,jk+1(xi + yj) = 1i<jk+1(xj xi)(yj yi) 1i,jk+1(xi + yj)

as we needed to show. □

39. [M23] Show that the inverse of a combinatorial matrix is a combinatorial matrix with the entries bij = (y + δij(x + ny))x(x + ny).

Proposition. [y + δijx]n1 = y+δij(x+ny) x(x+ny) n.

Proof. For

An = [aij]n = [y+δijx]n = y + x y y y y + x y y y y + x n,

we must show that

[aij]n1 = y + δij(x + ny) x(x + ny) n.

Let In = [δij]n and Jn = [1]n so that An = yJn + xIn, and note that Jn2 = nJn. Then

An(yJn + xIn) = x2I n ny2J n An(yJn + xIn) + ny2J n = x2I n An(yJn + xIn) + nyInyJn = x2I n An(yJn + xIn) + nyInyJn + xnyIn = x2I n + xnyIn An(yJn + xIn) + nyIn(yJn + xIn) = x2I n + xnyIn An(yJn + xIn) + nyInAn = x(x + ny)In An(yJn + xIn + nyIn) = x(x + ny)In An((x + ny)In yJn) = x(x + ny)In An (x + ny)In yJn x(x + ny) = In.

That is, for

Bn = (x + ny)In yJn x(x + ny) = (x + ny)[δij]n y[1]n x(x + ny) = (x + ny)δij y x(x + ny) n = y + δij(x + ny) x(x + ny) n,

AnBn = In, or equivalently, that

An1 = y + δij(x + ny) x(x + ny) n

as we needed to show. □

40. [M24] Show that the inverse of Vandermonde’s matrix is given by

bij =( 1k1<<knjn k1,,knji (1)j1x k1xknj)xi 1kn ki (xkxi).

Don’t be dismayed by the complicated sum in the numerator—it is just the coefficient of xj1 in the polynomial (x1 x)(xn x)(xi x).

Proposition. [xji]n1 = 1k1<<knjn k1,,knji (1)j1xk1xknj xi 1kn ki (xk xi).

Proof. Let

An = [aij]n = [xji] n = x1 x2 xn x12x22xn2 x 1nx 2nx nn n.

We must show that

[aij]n1 = 1k1<<knjn k1,,knji (1)j1x k1xknjxi 1kn ki (xk xi)n.

Let [bij]n = [aij]n1, so that by the definition of inverse and matrix multiplication, we have

[bij]n[aij]n = 1knbikakj n = 1knbikxjk n = [δij]n.

We require

1knbikxjk = δ ij,

or equivalently, given a polynomial Pi(x) = 1knbikxk for arbitrary x, we require

Pi(x) = δij

with given data points (δi1,x1),(δi2,x2),,(δin,xn). By polynomial interpolation, expanding the polynomial to an initial but trivial (n + 1)th variable x0 = 0, with bi0 = b0j = 0, in order to obtain a complete set of differences, we have

Pi(x) = 0kn δik 0mn mk x xm xk xm = 0kn k=i δik 0mn mk x xm xk xm + 0kn ki δik 0mn mk x xm xk xm = δii 0mn mi x xm xi xm + 0 = 0kn ki x xk xi xk = x x0 xi x0 1kn ki x xk xi xk = x xi 1kn ki xk x xk xi.

For x = xj, we have

1knbikxjk = P i(xj) = xj xi 1kn ki xk xj xk xi = δij.

What is left is to find the coefficients bij. By de Moivre’s work1 , we may do so.

The real and different roots of Pi(x) are exactly those xr where ri, since in such a case, we have

Pi(xr) = xr xi 1kn ki xk xr xk xi = (xrxr)xr xi 1kn ki kr xk xr xk xi = 0.

Let these roots be denoted by xr1,xr2,,xrn.

Since matrix multiplication with inverse is commutative, We also have

1knbikxjk = 1knxkib kj = δij.

By de Moivre’s identities, with our trivially expanded polynomial,

δij = 0knxkib kj bkj = 1mn(1)m 1r1<<rmn r1,,rmi xr1xrmδ(nm)jxk 1mn mk (xk xm)

we then have, since δ(nk)j requires j = n k and since n j and j 1 have opposite parity, that

bij = 1kn(1)k 1r1<<rkn r1,,rki xr1xrkδ(nk)jxi 1kn ki (xi xk) = (1)nj 1r1<<rnjn r1,,rnji xr1xrnjxi 1kn ki (xi xk) = (1)(1)j1 1r1<<rnjn r1,,rnji xr1xrnj(1)xi 1kn ki (xk xi) = 1r1<<rnjn r1,,rnji (1)j1x r1xrnjxi 1kn ki (xk xi).

Therefore

[aij]n1 = 1k1<<knjn k1,,knji (1)j1x k1xknjxi 1kn ki (xk xi)n

as we needed to show. □

______________________________________________________________________________________________________________________________

[A. de Moivre, The Doctrine of Chances, 2nd edition (London: 1738), 197–199.]

41. [M26] Show that the inverse of Cauchy’s matrix is given by

bij =( 1kn(xj+yk)(xk+yi))(xj+yi)( 1kn kj (xjxk))( 1kn ki (yiyk)).

Let

An = [aij]n = [1(xi+yi)]n = 1(x1 + y1)1(x1 + y2)1(x1 + yn) 1(x2 + y1)1(x2 + y2)1(x2 + yn) 1(xn + y1)1(xn + y2)1(xn + yn) n,

We must show that

[aij]n1 = 1kn(xj + yk)(xk + yi)(xj + yi) 1kn kj (xj xk) 1kn ki (yi yk)n.

But, by the definition of inverse,

[aij]n1 = [cofactor (aij)]nT det [aij]n = [cofactor (aji)]n 1u<vn(xv xu)(yv yu) 1u,vn(xu + yv) = (1)j+i det minor ([a]n,j,i) 1u<vn(xv xu)(yv yu) 1u,vn(xu + yv) = (1)j+i 1u<vn uj,ui vj,vi (xv xu)(yv yu) 1u,vn uj vi (xu + yv) 1u<vn(xv xu)(yv yu) 1u,vn(xu + yv) = (1)j+i 1u<vn uj vj (xu xv) 1u<vn ui vi (yu yv) 1u,vn uj vi (xu + yv) 1u<vn(xu xv) 1u<vn(yu yv) 1u,vn(xu + yv) = (1)j+i 1u,vn(xu + yv) 1u,vn uj vi (xu + yv) 1u<vn uj vj (xu xv) 1u<vn(xu xv) 1u<vn ui vi (yu yv) 1u<vn(yu yv) = (1)j+i 1u,vn u=jv=i (xu + yv) 1 1u<vn u=jv=j (xu xv) 1 1u<vn u=iv=i (yu yv) = (1)j+i(x j + yi) 1un uj (xu + yi) 1vn vi (xj + yv) 1 1uj1(xu xj) j+1vn(xj xv) 1 1ui1(yu yi) i+1vn(yi yv) = (1)j+i(x j + yi) 1un uj (xu + yi) 1vn vi (xj + yv) (1)j1 1un uj (xj xu) (1)i1 1vn vi (yi yv) = (1)2(j+i1)(x j + yi) 1un uj (xu + yi) 1un uj (xj xu) 1vn vi (xj + yv) 1vn vi (yi yv) = (xj + yi)2 1un uj (xu + yi) 1vn vi (xj + yv) (xj + yi) 1un uj (xj xu) 1vn vi (yi yv) = 1un(xu + yi) 1vn(xj + yv) (xj + yi) 1un uj (xj xu) 1vn vi (yi yv) = 1kn(xj + yk)(xk + yi) (xj + yi) 1kn kj (xj xk) 1kn ki (yi yk)

as we needed to show.

42. [M18] What is the sum of all n2 elements in the inverse of the combinatorial matrix?

For the inverse of the combinatorial matrix

1 y + δijxn1 = y + δij(x + ny) x(x + ny) n

the sum of all n2 elements is simply

1in 1jn y + δij(x + ny) x(x + ny) = 1in 1jn ji y y + x + ny x(x + ny) = 1in y(n 1) y + x + ny x(x + ny) = 1in ny + y y + x + ny x(x + ny) = 1in x x(x + ny) = 1in 1 x + ny = n x + ny.

43. [M24] What is the sum of all n2 elements in the inverse of Vandermonde’s matrix? [Hint: Use exercise 33.]

For the inverse of the Vandermonde matrix

xji n1 = 1k1<<knjn k1,,knji xk1xknj xi 1kn ki (xk xi)n = bij n,

we want to find the sum of all n2 elements 1i,jnbij.

In the case that any xκ = 1, 1 κ n, from exercise 40, we have that

[bij]n[xji] n = 1knbikxjk n = [δij]n

or equivalently that

1jnbij = δiκ 1in 1jnbij = 1 1i,jnbij = 1 (1 1xκ) 1i,jnbij = 1 1kn(1 1xk).

Otherwise, in the case that xκ1, 1 κ n, again from exercise 40, given Pi(x) = 1knbikxk for x = 1 we have that

1i,jnbij = 1inPi(x) = 1in x xi 1kn ki xk x xk xi

or equivalently, for x = 1, that

1inPi(1) = 1in 1 xi 1kn ki xk 1 xk xi = 1in xi 1 xi 1 1 xi 1kn ki xk 1 xk xi = 1in (xi 1) 1kn ki (xk 1) xi(xi 1) 1kn ki (xk xi) = 1in 1kn(xk 1) xi(xi 1) 1kn ki (xk xi) = 1kn(xk 1) 1in 1 xi(xi 1) 1kn ki (xk xi) = 1kn(1 xk) 1in 1 xi(xi 1) 1kn ki (xi xk) = 1kn(xk 1) 1in 1 xi 1kn ki (xi xk) 1in 1 (xi 1) 1kn ki (xi xk).

Also, from exercise 33, we have for an extrapolated and distinct x0 = 1 and for r = 0 that

0in 1 0jn ji (xi xj) = 1 1jn(1 xj)+ 1in 1 (xi 1) 1jn ji (xi xj) = 1

or equivalently that

1in 1 (xi 1) 1jn ji (xi xj) = 1 1 1jn(1 xj).

Similarly, if x0 = 0

1in 1 xi 1jn ji (xi xj) = 1 1 1jn(xj).

This yields

1i,jnbij = 1inPi(1) = 1kn(xk 1) 1in 1 xi 1kn ki (xi xk) 1in 1 (xi 1) 1kn ki (xi xk) = 1kn(xk 1) 1 1 1jn(xj) 1 1 1jn(1 xj) = 1kn(xk 1) 1 1jn(xj) 1 1jn(1 xj) = 1kn(xk 1) 1jn(xj) 1kn(xk 1) 1jn(1 xj) = 1kn(1 xk) 1jn(1 xj) 1kn(xk 1) 1jn(xj) = 1 1kn(xk 1) 1jn(xj) = 1 1knxk 1 xk = 1 1kn(1 1xk).

Therefore, in all cases,

1i,jnbij = 1 1kn(1 1xk).

44. [M26] What is the sum of all n2 elements in the inverse of Cauchy’s matrix?

For the inverse of the Cauchy matrix

[aij]n1 = [b ij]n = 1kn(xj + yk)(xk + yi) (xj + yi) 1kn kj (xj xk) 1kn ki (yi yk)n

we want to find the sum of all n2 elements 1i,jnbij.

But

1inbij = 1in 1kn(xj + yk)(xk + yi) (xj + yi) 1kn kj (xj xk) 1kn ki (yi yk) = 1in 1kn(xj + yk) 1kn(xk + yi) (xj + yi) 1kn kj (xj xk) 1kn ki (yi yk) = 1kn(xj + yk) 1kn kj (xj xk) 1in 1kn(xk + yi) (xj + yi) 1kn ki (yi yk) = 1kn(xj + yk) 1kn kj (xj xk) 1in 1kn ki (xk + yi) 1kn ki (yi yk) = 1kn(xj + yk) 1kn kj (xj xk)

since in general

1in 1kn1(yi zk) 1kn ki (yi yk) = 1

from exercise 34.

Then, for some polynomial P(x) of order n 2 and by exercise 33

1i,jnbij = 1in 1kn(xj + yk) 1kn kj (xj xk) = 1in xjn + 1knyk xjn1 + P(xj) 1kn kj (xj xk) = 1in xjn 1kn kj (xj xk) + 1knyk 1in xjn1 1kn kj (xj xk) + 1in P(xj) 1kn kj (xj xk) = 1knxk + 1knyk + 0 = 1kn(xk + yk).

45. [M25] A Hilbert matrix, sometimes called an n × n segment of the (infinite) Hilbert matrix, is a matrix for which aij = 1(i + j 1). Show that this is a special case of Cauchy’s matrix, find its inverse, show that each element of the inverse is an integer, and show that the sum of all elements of the inverse is n2. [Note: Hilbert matrices have often been used to test various matrix manipulation algorithms, because they are numerically unstable, and they have known inverses. However, it is a mistake to compare the known inverse, given in this exercise, to the computed inverse of a Hilbert matrix, since the matrix to be inverted must be expressed in rounded numbers beforehand; the inverse of an approximate Hilbert matrix will be somewhat different from the inverse of an exact one, due to the instability present. Since the elements of the inverse are integers, and since the inverse matrix is just as unstable as the original, the inverse can be specified exactly, and one should try to invert the inverse. The integers that appear in the inverse are, however, quite large.] The solution to this problem requires an elementary knowledge of factorials and binomial coefficients, which are discussed in Sections 1.2.5 and 1.2.6.

The Hilbert matrix, defined as

An = [aij]n = 1 i + j 1n = 1 1 2 1 n 1 2 1 3 1 n+1 1 n 1 n+1 1 2n1 n

is a special case of a Cauchy matrix Cn = [cij]n = 1 xi+yj n with xi = i and yj = j 1.

As such, its inverse [bij]n = 1 i+j1 n1 has elements given by

bij = 1kn(j + k 1)(k + i 1) (j + i 1) 1kn kj (j k) 1kn ki (i k) = 1kn(k + i 1) 1kn(k + i 1) (j + i 1) 1kj kj (j k) jkn kj (1)(k j) 1ki ki (i k) ikn ki (1)(k i) = (j + n 1)! (j 1)! (i + n 1)! (i 1)! (j + i 1)(j 1)! 1! (1)nj+1(n j)! 1! (i 1)! 1! (1)ni+1(n i)! 1! = (j + n 1)!(i + n 1)! (1)2nij+2(j + i 1)(j 1)!(i 1)!(j 1)!(n j)!(i 1)!(n i)! = (j + n 1)!(i + n 1)! (1)(i+j)(j + i 1)((j 1)!(i 1)!)2(n j)!(n i)! = (1)i+j(i + n 1)!(j + n 1)! (i + j 1)(i 1)!2(j 1)!2(n i)!(n j)!.

It is clear that bij is an integer, as it may be cast into binomial coefficients as

bij = (1)i+j(i + n 1)!(j + n 1)! (i + j 1)(i 1)!2(j 1)!2(n i)!(n j)! = (1)i+j 1 (j 1)!(i 1)! (i + n 1)! (i 1)! (j + n 1)! (j + i 1)(n i)! 1 (n j)!(j 1)! = (1)i+jj (i + j 2)! (j 1)!(i 1)! (i + n 1)! n!(i 1)! (j + n 1)! (j + i 1)!(n i)! n! (n j)!j! = (1)i+jj (i + j 2)! ((i + j 2) (i 1))!(i 1)! (i + n 1)! ((i + n 1) (i 1))!(i 1)! (j + n 1)! ((j + n 1) (n i))!(n i)! n! (n j)!j! = (1)i+jji + j 2 i 1 i + n 1 i 1 j + n 1 n i n j .

From exercise 44, the sum of the elements of the inverse is simply

1i,jnbij = 1kn(k + k 1) = 2 1knk 1kn1 = 2n(n + 1) 2 n = n2 + n n = n2.

______________________________________________________________________________________________________________________________

[For further information, see J. Todd, J. Research Nat. Bur. Stand. 65 (1961), 19–22; A. Cauchy, Exercices d’analyse et de physique mathématique 2 (1841), 151–159.]

46. [M30] Let A be an m × n matrix, and let B be an n × m matrix. Given that 1 j1,j2,,jm n, let Aj1j2jm denote the m × m matrix consisting of columns j1,,jm of A, and let Bj1j2jm denote the m × m matrix consisting of rows j1,,jm of B. Prove the Binet-Cauchy identity

det (AB) = 1j1<j2<<jmn det (Aj1j2 jm)det (Bj1j2 jm).

(Note the special cases: (i) m = n, (ii) m = 1, (iii) B = AT, (iv) m > n, (v) m = 2.)

Proposition. det (AB) = 1j1<j2<<jmn det (Aj1j2 jm)det (Bj1j2 jm).

Proof. Let A be an m × n matrix, and let B be an n × m matrix. Given that 1 j1,j2,,jm n, let Aj1j2jm denote the m × m matrix consisting of columns j1,,jm of A, and let Bj1j2jm denote the m × m matrix consisting of rows j1,,jm of B. We must show that

det (AB) = 1j1<j2<<jmn det (Aj1j2 jm)det (Bj1j2 jm).

Let

𝜖(k1,,km) = sign ( 1i<jm(kj ki))

for

sign (x) = [x > 0] [x < 0]

be the Levi-Civita function so that in general

det ([cij]n) = 1i1,,inn𝜖(i1,,in) 1inai,ii;

and so that if (k1,,km) and (l1,,lm) are identical except that ki = lj and kj = li, so that 𝜖(k1,,km) and 𝜖(l1,,lm), then in general

det (Bk1 km) = 𝜖(k1, ,km)det (Bj1 jm)

if j1 jm are the numbers k1,,km rearranged into nondecreasing order.

Then

det (AB) = 1l1,,lmm𝜖(l1,,lm) 1kna1kbkl1 1knamkbklm = 1k1,,kmna1k1amkm 1l1,,lmm𝜖(l1,,lm)bk1l1bkmlm = 1k1,,kmna1k1amkm det (Bk1 km) = 1k1,,kmn𝜖(k1,,km)a1k1amkm det (Bj1 jm) = 1j1jmn det (Aj1 jm)det (Bj1 jm)

as we needed to show.

Note that if m = n, we have the usual identity for square matrices

det (AB) = det (A)det (B);

if m = 1, we have the dot product

det (AB) = 1kna1kbk1 = A B;

if B = AT, we have the square

det (AB) = det (AAT) = 1j1jmn det (Aj1 jm)2;

and if m = 2, we have the first nontrivial case of the identity

det (AB) = 1j1j2n det (Aj1j2)det (Bj1j2).

______________________________________________________________________________________________________________________________

[J. de l’École Polytechnique 9 (1813), 280–354; 10 (1815), 29–112. Binet and Cauchy presented their papers on the same day in 1812.]

47. [M27] (C. Krattenthaler.) Prove that

det (x + q2)(x + q3)(x + p1)(x + q3)(x + p1)(x + p2) (y + q2)(y + q3)(y + p1)(y + q3)(y + p1)(y + p2) (z + q2)(z + q3)(z + p1)(z + q3)(z + p1)(z + p2) = (x y)(x z)(y z)(p1 q2)(p1 q3)(p2 q3).

and generalize this equation to an identity for an n × n determinant in 3n 2 variables x1,,xn,p1,,pn1,q2,,qn. Compare your formula to the result of exercise 38.

We may prove the more general equation.

Proposition. det 1kj1(xi + pk) j+1kn(xi + qk)n = 1i<jn(xi xj)(pi qj).

Proof. Let

An = [aij]n = 1kj1(xi + pk) j+1kn(xi + qk)n = 2kn(x1 + qk)(x1 + p1) 3kn(x1 + qk) 1kn1(x1 + pk) 2kn(x2 + qk)(x2 + p1) 3kn(x2 + qk) 1kn1(x2 + pk) 2kn(xn + qk)(xn + p1) 3kn(xn + qk) 1kn1(xn + pk) n.

We must show that

det [aij]n = 1i<jn(xi xj)(pi qj).

Let

aij = ai1 if j = 1 aij ai,j1if 2 j n

so that det [aij]n = det [aij]n where

aij = 1kj1(xi + pk) j+1kn(xi + qk) if j = 1 (pj1 qj) 1kj2(xi + pk) j+1kn(xi + qk)if 2 j n

since for 2 j n

aij = 1kj1(xi + pk) j+1kn(xi + qk) 1kj2(xi + pk) jkn(xi + qk) = (xi + pj1) 1kj2(xi + pk) j+1kn(xi + qk) 1kj2(xi + pk) (xi + qj) j+1kn(xi + qk) = ((xi + pj1) (xi + qj)) 1kj2(xi + pk) j+1kn(xi + qk) = (pj1 qj) 1kj2(xi + pk) j+1kn(xi + qk).

Let

qij = 1 if j = 1 pj1 qjif 2 j n

so that

[aij] n = ([qij]n[bij]nT)T

and that

det [aij] n = det (([qij]n[bij]nT)T) = det ([qij]n[bij]nT) = det [qij]n det ([bij]nT) = 2kn(pk1 qk)det ([bij]nT) = 2kn(pk1 qk)det [bij]n

where

[bij] = 1kj2(xi + pk) j+1kn(xi + qk)n.

Repeating this transformation, each time over fewer columns to factor out (pj2 qj1), (pj3 qj2), etc., we eventually have

det [aij]n = 1i<jn(pi qj)det [bij]n

for

[bij]n = j+1kn(xi + qk) n.

Then let

bij = bij (qj+1)bi,j+1if 1 j n 1 b in if j = n

so that det [bij]n = det [bij]n where

bij = xi j+2kn(xi + qk)if 1 j n 1 xi if j = n

since for 1 j n 1

bij = j+1kn(xi + qk) (qj+1) j+2kn(xi + qk) = (xi + qj+1) j+2kn(xi + qk) (qj+1) j+2kn(xi + qk) = ((xi + qj+1) (qj+1)) j+2kn(xi + qk) = xi j+2kn(xi + qk).

Repeating this transformation also, each time over fewer columns to factor out qj+2, qj+3, etc., we eventually have

det [aij]n = 1i<jn(pi qj)det [bij]n = 1i<jn(pi qj)det [bij] n

for

[bij] n = xinj n.

Let

cij = b1j if i = 1 bij b1jif 2 i n

and

cij = c1j x1c1(j+1)if i = 1,1 j n 1 c1n if i = 1,j = n cij x1ci(j+1)if 2 i n,1 j n 1 cin if 2 i n,j = n n,

so that det [bij] = det [cij ] where

cij = 0 if i = 1,1 j n 1 1 if i = 1,j = n xinj1(xi x1)if 2 i n,1 j n 1 0 if 2 i n,j = n n,

since:

c1j = b 1j = x 1nj i = 1,1 j n 1 c1n = b 1n = x 10 = 1 i = 1,j = n cij = b ij b 1j = x inj x 1nj 2 i n,1 j n 1 cin = b ij b 1j = x i0 x 10 = 0 2 i n,j = n n c1j = x 1nj x 1x1nj1 = 0 i = 1,1 j n 1 c1n = x 10 = 1 i = 1,j = n cij = x inj x 1nj x 1(xinj1 x 1nj1) = x inj1(x i x1)2 i n,1 j n 1 cin = x i0 x 10 = 0 2 i n,j = n n

Let

rij = xi+1 x1if i = j,1 i,j n 1 0 otherwise

so that minor ([b]n,1,n) = [rij]n1 minor ([c ]n,1,n) and det [rij]n1 = 1kn1(xk+1 x1). Then

det [cij ] n = 1incin cofactor (c in ) = c1n cofactor (c 1n ) + 2incin cofactor (c in ) = (1)(1)1+n det minor ([c ] n,1,n) + 0 = (1)n+1 det minor ([c ] n,1,n) = (1)n+1 det [r ij]n1 det (minor ([b] n,1,n)) = (1)n+1 det [r ij]n1 det minor ([b] n,1,n) = (1)n+1 1kn1(xk+1 x1)det minor ([b] n,1,n) = 1kn1(xk+1 x1)det minor ([b] n,1,n)

since a product of n + 1 signs will cancel with n 1 (i.e., (1)n+1(1)n1 = (1)2n = 1). This recursive identity allows us to prove by mathematical induction on n that

det [bij] k = 1i<jk(xi xj).

Therefore

det An = det [aij]n = det 1kj1(xi + pk) j+1kn(xi + qk)n = 1i<jn(pi qj)det [bij]n = 1i<jn(pi qj)det [bij] n = 1i<jn(pi qj) xinj n = 1i<jn(pi qj) 1i<jk(xi xj) = 1i<jn(xi xj)(pi qj)

as we needed to show. □

______________________________________________________________________________________________________________________________

[Manuscripta Math. 69 (1990), 177–178.]