Solutions to Chapter 2

Solution to Exercise 2.26. Suppose for a contradiction that R = N1 N2 where N1 , N2 are nonzero ideals of R. If a N1 and b N2 are any nonzero elements, then ab is a nonzero element of N1 N2 , which gives the contradiction. Note that we only used here the fact that R was an integral domain, not that it was a PID. If R is not an integral domain, it can be decomposable as a module over itself: for example, Exercise 1.108 showed that Z6 is decomposable as a Z-module, and the decomposition Z6 = {0, 3} {0, 2, 4} is equally valid in the category of Z6 -modules.

Solution to Exercise 2.27. Let M be a nitely-generated R-module. Choose a generating set {x1 , , xn } of M ; as seen in Proposition 1.71, this gives rise to a surjective R-module homomorphism : Rn M . If N is any submodule of M , then 1 (N ) is a submodule of Rn . But by Theorem 2.20, 1 (N ) must have some nite basis {y1 , , ym } where m n; since is surjective, we have N = (1 (N )), so N is generated by {(y1 ), , (ym )}. So we have shown that every submodule of a nitelygenerated R-module is nitely-generated. (As observed in Exercise 1.60, this is not true for the ring F [x1 , x2 , ]. The commutative rings R for which it is true are called noetherian after the German algebraist Emmy Noether.)

Solution to Exercise 2.29. Let N be the ideal of R generated by x and y ; as noted in Example 2.9, N is not principal. N is certainly a nitelygenerated torsion-free R-module, but it is not free, for the reason mentioned in Exercise 1.81: a basis of N would have to consist of at least two elements, 211

212

Modules and Group Representations

but any two nonzero elements r1 , r2 N satisfy the linear dependence r2 r1 + (r1 )r2 = 0.

Solution to Exercise 2.62.

(i) According to the procedures given in Section 2.3, we should consider the matrix 1 1 5 . Its determinant is 10, which implies that its 1 1 3 columns are linearly independent, so the set itself is an acceptable basis of the submodule in question. (Note that this submodule is not the whole of Z3 : for that, the determinant would have had to be 1). But supposing that we dont know the determinant but instead want to put the matrix in column-echelon form, the rst step is to right1 0 4 1 3 0 2 5 5 . Applying the multiply by 1 2 0 , which produces
0 2 3 4

column operation right-multiply by

basis for the submodule is {(1, 2, 2), (0, 1, 5), (0, 0, 10)}. (ii) Right-multiplying
x x+1 x 2 x x 2 1

3 1 0 0 C3 C3 4C1 , this becomes 2 5 3 2 5 5 10 0 1 0 0 0 1 3 , which produces 2 1 0 . 0 2 5 2 5 10 0 1

. We then So another

by the invertible matrix

1 x 1 1 x

, we

1 0 , which implies that the submodule has rank 1 and basis get x 10 {(1, x 1)}.

(iii) We have a chain of column equivalences: 2x 1 x x x x2 x 2 3x 1 x + 3x x + 1 1 x x x x2 x C1 C1 2C3 2 x 3 x + 3x x + 1 1 0 0 x 0 x x2 . C2 C2 +xC1 x 3 0 x2 2x + 1 C3 C3 +xC1 We could obviously put the latter matrix in column-echelon form by swapping the second and third columns. So a basis is {(1, x, x 3), (0, x x2 , x2 2x + 1)}.

CHAPTER 6. REAL REPRESENTATIONS Solution to Exercise 2.63.

213

(i) Clearly x + xy + N = (1 + y )(x + N ) belongs to the ideal of R/N generated by x + N , and x + N = x xy 2 + N = (1 y )(x + xy + N ) belongs to the ideal of R/N generated by x + xy + N , so x + N and x + xy + N generate the same ideal of R/N . We must now nd which elements of R/N are invertible. But if p + N is invertible in R/N , its image in the quotient ring R/Ry = F [x] must also be invertible, which means that the constant term of p is nonzero and p contains no xi terms for i 1. Similarly p contains no y i terms for i 1, so p + N must be of the form a + bxy + N where a, b F and a is nonzero. Conversely, any element of this form is invertible, because (a + bxy + N )(a1 a2 bxy + N ) = 1 a2 b2 x2 y 2 + N = 1 + N. However, (x + N )(a + bxy + N ) = ax + bx2 y + N = ax + N cannot equal x + xy + N , so x + xy + N cannot be obtained from x + N by multiplying by an invertible element of R/N . (This phenomenon of having nonassociate elements generating the same ideal couldnt happen in an integral domain. It can also be shown that it cant happen in a ring which is a product of quotients of PIDs, which is why this example had to be so far-fetched.) (ii) Clearly R{x + xy, y 2 } R{x, y 2 }, and the other direction follows from q x = (1 y )(x + xy ) + xy 2 . Suppose that Y = ( p r s ) Mat2 (R) was invertible and satised (x + xy y 2 ) = (x y 2 )Y . Then det(Y ) is an invertible element of R; in other words, ps qr is a nonzero constant. The matrix equation is equivalent to x + xy = xp + y 2 r, y 2 = xq + y 2 s. The second of these equations forces q = y 2 q , s = 1 xq for some q R. The rst forces p = 1 + y + y 2 p , r = xp for some p R. We then have ps qr = (1 + y + y 2 p )(1 xq ) + xy 2 p q = 1 + y + y 2 p xq xyq , which clearly has a nonzero y term, giving us the desired contradiction. (As noted in Remark 2.61, this phenomenon cant happen in a principal ideal domain.)

214 Solution to Exercise 2.92.

Modules and Group Representations

(i) If (a) and (b) are 1 1 matrices over R, then they are equivalent if and only if b = xay for some invertible x, y R; in other words, if and only if a and b are associates in R. So an example is A = (1) and B = (2), since 1 and 2 are associates in Q but not in Z. (ii) This time there is no hope of such a simple example, because for 1 1 (or, more generally, 1 m) matrices there is no dierence between being column-equivalent over Z and being equivalent over Z. It is also easy to see that there cannot be an example using n 1 matrices, so the smallest possible size is 2 2. One example is A= 2 5 , B= 0 3 2 0 . 0 3

These are equivalent over Z because B can be obtained from A by applying the column operation C2 C2 C1 and the row operation R1 R1 R2 . They are also column-equivalent over Q because B can be obtained from A by the column operation C2 C2 5 2 C1 . But A and B are not column-equivalent over Z, i.e. there is no invertible integer matrix Y such that A = BY ; this follows from the fact that 1 5 2 B 1 A = has a non-integer entry. 0 1 (iii) Since A and B must have the same determinant, we need the determinant to have a repeated prime factor so that there is scope for A and B to have dierent invariant factors. One example is A= 1 1 , B= 1 5 2 2 . 10 8

These both have determinant 4 and trace 6, so they have the same characteristic polynomial, namely x2 6x + 4. As we will see later, the fact that this polynomial is irreducible over Q means that all matrices which have it as their characteristic polynomial are conjugate over Q; but in any case, we can see directly that B = X 1 AX where 10 3 X= . (There are many other choices of conjugating matrix; 0 1 in this choice X does have integer entries, but X 1 doesnt.) However, A and B are not equivalent over Z, because A has invariant factors 1, 4 while B has invariant factors 2, 2.

CHAPTER 6. REAL REPRESENTATIONS Solution to Exercise 2.93.

215

(i) We have a1 = gcd(6, 15) = 3 and a1 a2 = 6 15 = 90. So the invariant factors are 3 and 30. Alternatively, apply the method of ordering the prime factors as in Proposition 2.87.
0 0 0 (ii) Clearing the rst row with column operations gives 1 2 1 2 3 . We can then eliminate the 2 with R2 R2 2R1 and then the 2 and the 0 00 3 with column operations to get the normal-form matrix 1 0 1 0 0 . So the invariant factors are 1 and 1. In terms of minors, this expresses the fact that the 2 2 minors have no common factor (indeed, the determinant of the left-hand 2 2 submatrix is 1).

(iii) Algorithm 2.76 gives that this matrix is equivalent to

point it is best to diverge from the algorithm (which does something drastic since 2 15), and consider the minors of the simpler matrix. The gcd of the entries is obviously 1, so a1 = 1; the gcd of the 3 3 minors is 10, so a1 a2 a3 = 10; and then the divisibility requirement shows that a2 = 1, a3 = 10. (iv) The gcd of the entries is 1 and the determinant is 12, so the invariant factors must be 1, a2 , 12/a2 where a2 2 | 12. The only possibilities for a2 are 1 and 2, so we just need to determine whether all the 2 2 minors are even. But the second row is all even, so any submatrix involving that will have even determinant; and any 2 2 submatrix avoiding the second row will have all four entries odd, hence even determinant. So the invariant factors are 1, 2, 6. (Of course you can also get this via row and column operations.)

1 0 0 0

0 2 0 0

0 0 15 20

. At this

Solution to Exercise 2.94. (i) Following the method given by Proposition 2.87 for nding invariant factors of a diagonal matrix, we nd the irreducible factorizations: 1 + x = x0 (1 x)0 (1 + x)1 , x = x1 (1 x)0 (1 + x)0 , 1 x2 = x0 (1 x)1 (1 + x)1 . It follows that the invariant factors are 1, 1 + x, x(1 x)(1 + x).

216

Modules and Group Representations

(ii) We can see that the rst invariant factor a1 is 1 because the entries have no common factor. Calculating some 2 2 minors reveals that they have no common factor either, so a1 a2 is also 1. The determinant of the whole matrix is 3x3 + 4x2 3x, so the invariant factors are 1, 1, 3x3 + 4x2 3x. The fact that the determinant has no repeated irreducible factors forces the rst two invariant factors to be 1, so the calculation of minors was not actually required. (Of course you can also nd the invariant factors via row and column operations.)

Solution to Exercise 2.95. The gcd of the entries is quickly seen to be 1 + i (which divides 2 = i(1 + i)2 ). There are three 2 2 submatrices to consider, and their determinants are 12, 2 2i, and 4. The gcd of these is 2 2i = (1 + i)3 . So the invariant factors are 1 + i and (1 + i)2 . Solution to Exercise 2.96. (i) We know that XAY = diag(a1 , , amin{n,m} ) for some invertible matrices X, Y . So X (rA)Y = rXAY = diag(ra1 , , ramin{n,m} ), and we also have ra1 | ra2 | | ramin{n,m} , proving the claim. Alternatively, one can use the fact that the k k minors of rA are r k times the k k minors of A. (ii) Transposing the equation XAY = diag(a1 , , amin{n,m} ), we obtain Y t At X t = diag(a1 , , amin{n,m} )t = diag(a1 , , amin{n,m} ). Now the fact that X and Y are invertible implies that X t and Y t are also (for example, transposing XX 1 = 1n gives (X 1 )t X t = 1n ). So At is equivalent to diag(a1 , , amin{n,m} ), and hence has the same invariant factors as A. (iii) This is a bit of a trick question: if A is invertible, then det(A) is a unit in R, so its invariant factors are all units in R, and we may as well call them all 1. (Conversely, a square matrix whose invariant factors are all units has determinant which is a unit and is therefore invertible.) So the answer to the question is that the invariant factors of A1 (which of course is also invertible!) are all 1.

CHAPTER 6. REAL REPRESENTATIONS

217

(iv) It is a fact that for any A, B Matn (R), adj (AB ) = (adj B )(adj A). Thanks to the assumption of nonzero determinant, we only need the case of this where det(A) and det(B ) are nonzero; this case follows from the equation AB adj (AB ) = det(AB )1n = det(B ) det(A)1n = det(B )A(adj A) = A(det(B )1n )(adj A) = AB (adj B )(adj A), since the columns of AB must be linearly independent for det(AB ) = 0 to hold, and that means we can cancel AB from the left. It follows that if X Matn (R) is invertible, then so is adj X . Taking adjugates of the equation XAY = diag(a1 , , an ), we deduce that (adj Y )(adj A)(adj X ) = adj(diag(a1 , , an )) = diag(a2 a3 an , a1 a3 an , , a1 a2 an1 ). The latter diagonal matrix is not in normal form, but we can simply reverse the order of the rows and columns to make it so. So the answer is that the invariant factors of adj A are a1 a2 an1 , , a2 a3 an .

Solution to Exercise 2.97.

(i) There are plenty of examples to demonstrate this. One could consider 0 00 A = (1 0 0 ) and B = ( 0 1 ), so that a1 = b1 = 1, a2 = b2 = 0, whereas 0 0 AB = ( 0 0 ) has invariant factors 0, 0. (ii) Recall from Exercise 2.96(ii) that the invariant factors of the transpose matrix At are the same as those of A. Since (AB )t = B t At , taking transpose of all matrices interchanges the role of A and B in the question. So it suces to show that ak | ck for k = 1, 2, , n. Write DA , DB , DC for diag(a1 , a2 , . . . , an ) etc. Then A = XA DA YA and B = XB DB YB for some invertible XA , XB , YA , YB Matn (R), so
1 1 is equivalent to DC . ABYB DA YA XB DB = XA

All we need to take away from this is that c1 , , cn are the invariant factors of DA M for some M = (mij ) Matn (R). There are now (at least) three ways to proceed.

218

Modules and Group Representations Method 1. For 1 k n, consider any k k minor of DA M , say the determinant of the submatrix formed using rows i1 , , ik and columns j1 , , jk . By expanding this determinant along its nal row, we see that it is a multiple of aik times the gcd of all (k 1) (k 1) minors of DA M . (To make this true when k = 1, we need the convention that the determinant of the unique 0 0 matrix is 1.) We know that this gcd is c1 ck1 . Also aik is a multiple of ak (since ik k). So c1 ck1 ak divides every k k minor of DA M , and hence divides their gcd, which is c1 ck . So we have shown that c1 ck1 ak | c1 ck1 ck . If c1 ck1 = 0, we can conclude that ak | ck . If c1 ck1 = 0, then we must have ci = 0 for some 1 i k 1, and then the divisibility constraints imply that ck = 0, so ak | ck in this case also. Method 2. We have DA M = XDC Y for some invertible X, Y Matn (R). Renaming M as M Y , we can get rid of Y , and thus get DA M = XDC , which says that ai mij = xij cj for all i, j . Now suppose for a contradiction that ak ck . Then there must be an irreducible element p R and a positive integer such that p | ak but p ck . Since ak | ai for all i k, we have p | xij cj for all i k. Since cj | ck for all j k, we have p cj for all j k, and we conclude that p | xij for all i k, j k. But every term of det(X ) must contain some such xij , so this would mean p | det(X ), which is impossible because det(X ) is a unit. Method 3. The matrix DA M has the property well call Property K: that ak divides the kth row, for k = 1, , n. It suces to show that we can carry out invertible row and column operations which will bring the matrix into normal form while maintaining Property K all the time; at the end of the procedure, ak will divide the kth entry of the diagonal, which will be ck . Any column operations clearly maintain Property K, as do the operations of multiplying a row by a unit. A row operation Ri Ri + cRj for i < j also maintains Property K, because ai | aj . A row operation Ri Ri + cRj for i > j usually wont maintain Property K, but it clearly will in the special case that Rj is zero except for the (j, ) entry and the new Ri has a zero in the (i, ) entry. Similarly, a row swap usually wont maintain Property K, but it clearly will in the special case that the row with the smaller number is zero. It is easy to make a slight modication of Algorithm 2.76 so that we only ever use these special cases of the last two kinds of operations.

CHAPTER 6. REAL REPRESENTATIONS

219

Solution to Exercise 2.112. The presentation matrix A in this case is the 2 1 matrix ( a b ). The main point is that the sole invariant factor of this matrix is d (the gcd of the entries). We now have three cases. If d = 0 (i.e. a = b = 0), the module Rx + Ry has no torsion invariants, and its rank equals 2 rk(A) = 2, so it is isomorphic to R R = R/R0 R as claimed. If d is a unit, the module Rx + Ry has no torsion invariants, and its rank equals 2 rk(A) = 1, so it is isomorphic to R = R/R1 R as claimed. If d a nonzero non-unit, the module Rx + Ry has sole torsion invariant d, and its rank equals 2 rk(A) = 1, so it is isomorphic to R/Rd R.

Solution to Exercise 2.113.

(i) Clearly rk(Zn ) = 0 and the sole torsion invariant is n (except if n = 1, in which case there are no torsion invariants and the module is {0}). (ii) Again we are dealing with a torsion module, so the rank is 0. The torsion invariants can be obtained by factorizing 20, 140, and 108 and then ordering the powers of each prime: they are 22 = 4, 22 5 = 20, and 22 33 5 7 = 3780. (That is, M = Z4 Z20 Z3780 .) (iii) The presentation matrix is
2 3 4 1 1 5 1 1 3

. From the calculations in Ex-

ercise 2.62(i), it is easy to see that the invariant factors of this matrix are 1, 1, 10. So M has rank 3 3 = 0, and sole torsion invariant 10. (That is, M = Z10 .) (iv) The eld Zp is free of rank 1 as a module over itself, and there are no torsion invariants. (v) We have M = F [x]/F [x](x 3), so the rank is 0 and the sole torsion invariant is x 3. (vi) We have M = F [x]/F [x]x2 , so the rank is 0 and the sole torsion 2 invariant is x .

220

Modules and Group Representations

Solution to Exercise 2.114. By denition, the invariant factors of M in Rn are all 1 if and only if there is a basis {f1 , , fn } of Rn such that {f1 , , fr } is a basis of M for some 0 r n. If this is the case, then it is easy to see that the submodule generated by the remaining basis elements, namely R{fr+1 , , fn }, is complementary to M in Rn ; moreover, the images fr+1 + M, , fn + M form a basis for Rn /M . So (1) implies (2) and (3). If we assume the existence of a complementary submodule M , then knowing that both M and M must have nite bases, we can take the union of these to get a basis of M M = Rn ; also, Rn /M is isomorphic to M and hence free. So (2) implies (1) and (3). We saw that (3) implies (2) in Exercise 1.115. This gives more than enough implications to prove the result.

Solution to Exercise 2.115. First consider the special case where M is a free module, i.e. M is isomorphic to Rn for some n. By Theorem 2.98, there is some basis {f1 , , fn } of M such that {a1 f1 , , ar fr } is a basis of N , where r = rk(N ) and a1 , , ar are the invariant factors of N in M . As seen in the proof of Theorem 2.102, it then follows that M/N = R/Ra1 R/Rar R R, where the number of copies of R is n r . Hence rk(M/N ) = n r = rk(M ) rk(N ) as required. Now return to the general case where M is not necessarily free. Since M is nitely-generated, there is a surjective R-module homomorphism : Rn M for some n. Let N = 1 (N ), a submodule of Rn containing ker(); since is surjective, the restriction : N N is also surjective. Then by the First Isomorphism Theorem we have M = Rn / ker() and N = N / ker(), n and by the Third Isomorphism Theorem we have M/N = R /N . By the special case we have already proved, these isomorphisms tell us that rk(M ) = n rk(ker()), rk(N ) = rk(N ) rk(ker()), rk(M/N ) = n rk(N ). From these equations it obviously follows that rk(M/N ) = rk(M ) rk(N ).

Solution to Exercise 2.127. In part (i), the primary invariants are the prime powers in the factorization of n. In part (ii), the primary invariants are the prime-power factors of the torsion invariants 4, 20, 3780 (or equivalently

CHAPTER 6. REAL REPRESENTATIONS

221

of the original numbers 20, 140, 108): namely, 22 , 22 , 22 , 33 , 5, 5, and 7. In part (iii), the primary invariants are 2 and 5. In part (iv) there are no primary invariants. In part (v) the sole primary invariant is x 3, and in part (vi) the sole primary invariant is x2 .

Solution to Exercise 2.128. (i) Obviously G[k] contains the identity and is closed under taking ink k k verses. If G is abelian and g, h G[k] , then (gh)p = gp hp = 1, so gh G[k] also. This shows that G[k] is a subgroup. (ii) Let G be the group of symmetries of a square, a group of order 8 = 23 consisting of four reections and four rotations (counting the identity as a rotation through 0 degrees). If g is a reection about a line joining two corners and h is a reection about a line joining the midpoints of opposite sides, then g2 = h2 = 1, so g, h G[1] . However, gh is then a rotation through 90 degrees, so it has order 4 rather than 2, and gh G[1] . (iii) Note that Cpm is all of Cpm if k m, and otherwise is a subgroup isomorphic to Cpk . So we have |G[k] | = |Cpn1 | |Cpn2 | |Cpns | = pmin{k,n1 } pmin{k,n2 } pmin{k,ns } = pmin{k,n1 }+min{k,n2 }++min{k,ns } . If we now replace k by k + 1, the terms in the exponent which will change are those min{k, ni } where k < ni ; these will each increase by 1. So knowing all the numbers |G[k] | tells us, for every k = 0, 1, , n, how many of the ni s are strictly greater than k. The amount by which this number decreases as k 1 is replaced by k tells us how many of the ni s are equal to k, and that determines the sequence of ni s.
[k ] [k ] [k ] [k ]