Janson Diskriminanter
Janson Diskriminanter
SVANTE JANSON
1. Resultant
Definition 1.1. Let f (x) = an xn + · · · + a0 and g(x) = bm xm + · · · + b0 be
two polynomials of degrees (at most) n and m, respectively, with coefficients
in an arbitrary field F . Their resultant R(f, g) = Rn,m (f, g) is the element
of F given by the determinant of the (m + n) ⇥ (m + n) Sylvester matrix
Syl(f, g) = Syln,m (f, g) given by
0 1
an an 1 an 2 ... 0 0 0
B0 an an 1 ... 0 0 0C
B . .. .. .. .. .. C
B . C
B . . . . . .C
B C
B0 0 0 ... a1 a0 0C
B C
B0 0 0 ... a2 a1 a0 C
B C (1.1)
Bb m b m 1 b m 2 ... 0 0 0C
B C
B0 bm bm 1 ... 0 0 0C
B . .. .. .. .. .. C
B .. . . . . .C
B C
@0 0 0 ... b1 b 0 0A
0 0 0 ... b2 b 1 b0
Example. If n = 3 and m = 2,
a3 a2 a1 a0 0
0 a3 a2 a1 a0
R(f, g) = b2 b1 b0 0 0 .
0 b 2 b1 b 0 0
0 0 b2 b 1 b 0
In the exterior algebra over F [x], we thus have
(xm 1
f (x)) ^ (xm 2
f (x)) ^ · · · ^ f (x)
^ (xn 1
g(x)) ^ (xn 2
g(x)) ^ · · · ^ g(x)
= R(f, g)xn+m 1
^ xn+m 2
^ · · · ^ 1, (1.2)
which can be used as an alternative form of Definition 1.1.
Remark 1.2. Typically, one assumes in Definition 1.1 that n = deg(f )
and m = deg(g), i.e. that an 6= 0 and bm 6= 0; this implies that R(f, g)
is completely determined by the polynomials f and g (and it excludes the
case f = 0 or g = 0). It is, however, convenient to use the slightly more
general version above which also allows n and m to be regarded as given and
then R(f, g) is defined for all polynomials f and g of degrees deg(f ) n,
deg(g) m. (See for example Remarks 1.9 and 3.4.) In this case, we may
use the notation Rn,m (f, g) to avoid ambiguity, but usually we write just
R(f, g).
Remark 1.3. It is sometimes convenient to regard ai and bj is indetermi-
nates, thus regarding f and g as polynomials with coefficients in the field
F (an , . . . , a0 , bm , . . . , b0 ). Any formula or argument that requires an 6= 0
and bm 6= 0 then can be used; if this results in, for example, a polynomial
identity involving Rn,m (f, g), then this formula holds also if we substitute
any values in F for an , . . . , b0 .
The resultant is obviously a homogeneous polynomial of degree n + m
with integer coefficients in the coefficients ai , bj . More precisely, we have
the following. We continue to use the notations ai and bj for the coefficients
of f and g, respectively, as in Definition 1.1.
Theorem 1.4. Rn,m (f, g) is a homogeneous polynomial with integer coeffi-
cients in the coefficients ai , bj .
(i) Rn,m (f, g) is homogeneous of degree m in an , . . . , a0 and degree n in
bm , . . . , b 0 .
(ii) If ai and bi are regarded as having degree i, then Rn,m (f, g) is ho-
mogeneous of degree nm.
Proofs of this and other results in this section are given in Section 2.
Remark 1.5. If we write Rn,m (f, g) as a polynomial with integer coefficients
for any field with characteristic 0, such as Q or C, then the formula is
RESULTANT AND DISCRIMINANT OF POLYNOMIALS 3
valid (with the same coefficients) for every field F . (Because the coefficients
are given by expanding the determinant of Syln,m (f, g) and thus have a
combinatorial interpretation independent of F . Of course, for a field of
characteristic p 6= 0, the coefficients may be reduced modulo p, so they are
not unique in that case.)
The main importance of the resultant lies in the following formula, which
often is taken as the definition.
Theorem 1.6. Let f (x) = an xn +· · ·+a0 and g(x) = bm xm +· · ·+b0 be two
polynomials of degrees n and m, respectively, with coefficients in an arbitrary
field F . Suppose that, in some extension of F , f has n roots ⇠1 , . . . , ⇠n and
g has m roots ⌘1 , . . . , ⌘m (not necessarily distinct). Then
n Y
Y m
R(f, g) = am n
n bm (⇠i ⌘j ). (1.3)
i=1 j=1
Here and below, the roots of a polynomial are listed with multiple roots
repeated according to their multiplicities. Thus every polynomial of degree
n has n roots in some extension field (for example in an algebraically closed
extension). Combining Theorem 1.6 and Definition 1.1, we see that the
product in (1.3) lies in coefficient field F , and that it does not depend on
the choice of extension field.
Theorem 1.6 implies the perhaps most important result about resultants.
Corollary 1.7. Let f and g be two non-zero polynomials with coefficients
in a field F . Then f and g have a common root in some extension of F if
and only if R(f, g) = 0.
It is here implicit that R = Rn,m with n = deg(f ) and m = deg(g). Since
f and g have a common root in some extension of F if and only if they have
a common non-trivial (i.e., non-constant) factor in F , Corollary 1.7 can also
be stated as follows.
Corollary 1.8. Let f and g be two non-zero polynomials with coefficients
in a field. Then f and g have a common non-trivial factor if and only if
R(f, g) = 0. Equivalently, f and g are coprime if and only if R(f, g) 6= 0.
Remark 1.9. If n and m are fixed, we can (in the style of projective ge-
ometry) say that a polynomial f with deg(f ) n has n deg(f ) roots at
1, and similarly g has m deg(g) roots at 1. Thus f always has n roots
and g has m, in F1 [ {1} for some extension F1 . With this interpretation,
Corollary 1.7 holds for all polynomials with deg(f ) n and deg(g) m.
(Including f = 0 and g = 0 except in the trivial case n = m = 0.)
We have also the following useful formulas related to (1.3).
Theorem 1.10. Let f (x) = an xn + · · · + a0 and g(x) = bm xm + · · · + b0 be
two polynomials with coefficients in an arbitrary field F .
4 SVANTE JANSON
This formula (ignoring the middle expression) holds also if a = 0 (and then
simplifies to R1,m (c, g) = bm ( c)m ), for example by Remark 1.3.
Example 1.17. Let n 0 and let f (x) and g(x) be two polynomials of
degree n with coefficients in a field F . Further, let a, b, c, d 2 F .
Assume first that d 6= 0. Then, by Theorem 1.14 and Theorem 1.4,
Rn,n (af + bg, cf + dg) = Rn,n af + bg (b/d)(cf + dg), cf + dg
= Rn,n (a bc/d)f, cf + dg
n
= (a bc/d) Rn,n (f, cf + dg)
= (a bc/d)n Rn,n (f, dg)
= (ad bc)n Rn,n (f, g). (1.16)
The final formula holds in the case d = 0 too, for example by regarding d as
an indeterminate. We may write the result as
n
a b
Rn,n (af + bg, cf + dg) = Rn,n (f, g). (1.17)
c d
6 SVANTE JANSON
a2 a1 a0 0
0 a2 a1 a0
R(f, g) = = (a2 b0 b2 a 0 ) 2 (a2 b1 b2 a1 )(a1 b0 b1 a0 ).
b 2 b 1 b0 0
0 b 2 b1 b 0
RESULTANT AND DISCRIMINANT OF POLYNOMIALS 7
If n = m = 3,
a3 a2 a1 a0 0 0
0 a3 a2 a1 a0 0
0 0 a3 a2 a1 a0
R(f, g) = .
b 3 b2 b 1 b 0 0 0
0 b3 b 2 b 1 b 0 0
0 0 b3 b2 b1 b0
More generally, if m = n, then
an an 1 ... a0 0 0 ... 0
0 an ... a1 a0 0 ... 0
.. .. .. .. .. ..
. . . . . .
0 0 ... a n 1 an 2 0 ... a0
R(f, g) = .
b n bn 1 ... b0 0 0 ... 0
0 bn ... b1 b0 0 ... 0
.. .. .. .. .. ..
. . . . . .
0 0 ... bn 1 b n 2 0 ... b0
2. Proofs
We begin by noting that, as said above, (1.6) holds by a permutation of
the rows in (1.1).
and column equals Syln,m 1 (f, g). Hence, by expanding the determinant
Rn,m (f, g) along the first column,
Rn,m (f, g) = an Rn,m 1 (f, g).
Proof of Theorems 1.6 and 1.10. We prove these theorems together by in-
duction on n + m.
Assume that deg(fQ) = n and deg(g) = m. Then, Qmat least in some exten-
n
sion field, f (x) = an i=1 (x ⇠i ) and g(x) = bm j=1 (x ⌘j ), and (1.3) is
equivalent to both (1.4) and (1.5). Assume by induction that these formulas
hold for all smaller values of n + m (and all polynomials of these degrees).
Case 1. First, suppose 0 < n = deg(f ) m = deg(g). Divide g by f to
obtain polynomials q and r with g = qf + r and deg(r) < deg(f ) = n. Note
that
deg(q) = deg(qf ) deg(f ) = deg(g r) n = m n.
By Theorem 1.14,
Rn,m (f, g) = Rn,m (f, g qf ) = Rn,m (f, r). (2.1)
Case 1a. Suppose that r 6= 0 and let k := deg(r) 0. By Theorem 1.13
and the inductive hypothesis in the form (1.4),
n
Y n
Y
Rn,m (f, r) = am
n
k
Rn,k (f, r) = am
n
k k
an r(⇠i ) = am
n g(⇠i ),
i=1 i=1
since g(⇠i ) = q(⇠i )f (⇠i ) + r(⇠i ) = r(⇠i ), which verifies (1.4) and thus (1.3).
Case 1b. Suppose now that r = 0, so g = qf , but n > 0. Then
Syln,m (f, r) = Syln,m (f, 0) has the last n rows identically 0, so Rn,m (f, r) =
0, and Rn,m (f, g) = 0 by (2.1). Further, g(⇠1 ) = q(⇠1 )f (⇠1 ) = 0 so ⇠1 is a
root of g too, and the right hand side of (1.3) vanishes too. Hence, (1.3)
holds.
Case 2. Suppose that n = 0. As remarked in Subsection 1.1, R0,m (f, g) =
am
0 , which agrees with (1.3). (This includes the case n = m = 0 that starts
the induction.)
Case 3. Suppose that m = deg(g) < n = deg(f ). This is reduced to Case
1 or 2 by (1.6).
This completes the induction, and the proof of Theorem 1.6. It remains
to verify Theorem 1.10(i),(ii) also in the cases deg(g) < m and deg(f ) < n,
respectively. This follows by Theorem 1.13, as in the proof of Case 1a above,
or by Remark 1.3. ⇤
RESULTANT AND DISCRIMINANT OF POLYNOMIALS 9
Proof of Corollaries 1.7 and 1.8. Immediate from (1.3), using the fact on
common factors stated before Corollary 1.8. ⇤
Proof of Theorem 1.11. By the argument in Remark 1.3, we may assume
that deg(g) = m, so g has m roots in some extension of F , and then (1.7)
follows from (1.5). Similarly, (1.8) follows from (1.4). ⇤
Proof of Theorem 1.12. The Sylvester matrix Syln,m (f ⇤ , g ⇤ ) is obtained from
Sylm,n (g, f ) by reversing the order of both rows and columns, and thus they
have the same determinant. ⇤
Qm
Proof of Theorem 1.15. We have g(x) = bm j=1 (x ⌘j ) and thus
m
Y m
Y
g(y x) = bm (y x ⌘j ) = ( 1)m bm (x y + ⌘j ).
j=1 j=1
Proof of Theorem 1.18. We work in the exterior algebra over F (x), using
(1.2). Let D be the determinant in (1.18); thus
rn 1 (x) ^ · · · ^ r0 (x) = Dxn 1
^ · · · ^ x0 . (2.2)
For k n 1, xk g(x) rk (x) = qk (x)f (x) has degree n + m 1 and is
thus a linear combination of f (x), xf (x), . . . , xm 1 f (x); hence, using (2.2)
and (1.2) (with g(x) replaced by 1),
(xm 1
f (x)) ^ · · · ^ f (x) ^ (xn 1
g(x)) ^ · · · ^ g(x)
= (xm 1
f (x)) ^ · · · ^ f (x) ^ rn 1 (x) ^ · · · ^ r0 (x)
m 1 n 1
= (x f (x)) ^ · · · ^ f (x) ^ Dx ^ · · · ^ x0
= DRn,m (f, 1)xn+m 1
^ · · · ^ 1.
Consequently, using (1.2) again,
Rn,m (f, g) = DRn,m (f, 1).
Finally, by Theorem 1.13, or by (1.4),
Rn,m (f, 1) = am m
n Rn,0 (f, 1) = an . ⇤
Proof of Theorem 1.19. Note first that Syln,m (f, g) and Sylm,n (g, f ) have
the same rank and corank, so we may interchange f and g. We may thus
assume n m.
In this case, we may as in the proof of Theorem 1.14 for any polynomial
q with deg(q) n m obtain Syln,m (f qg, g) from Syln,m (f, g) by row
10 SVANTE JANSON
operations that do not change the rank and corank. In particular, we may
replace f by the remainder r obtained when dividing f by g. Then deg(r) <
m n and the first column of Syln,m (r, g) has a single non-zero element,
bm in row m + 1. We may thus delete the first column and the m + 1:th
row without changing the corank, and this yields Syln 1,m (r, g). (Cf. the
proof of Theorem 1.13.) Repeating, we see that if r 6= 0 and k = deg(r),
then Syln,m (f, g) has the same corank as Sylk,m (r, g) and Sylm,k (g, r). We
repeat from the start, by dividing g by r and so on; this yields the Euclidean
algorithm for finding the GCD h, and we finally end up with the Sylvester
matrix Sylk,l (0, h), for some k 0 and l = deg(h), which evidently has
corank l since the first l rows are 0 and the last k are independent, as is
witnessed by the lower left k ⇥ k minor which is triangular.
(Alternatively, Theorem 1.19 follows easily from Theorem 1.20.) ⇤
3. Discriminant
Several di↵erent normalizations of the discriminant of a polynomial are
used by di↵erent authors, di↵ering in sign and in factors that are powers of
the leading coefficient of the polynomial. One natural choice is the following.
Definition 3.1. Let f be a polynomial of degree n 1 with coefficients
in an arbitrary field F . Let F1 be an extension of F where f splits, and
let ⇠1 , . . . , ⇠n be the roots of f in F1 (taken with multiplicities). Then the
(normalized) discriminant of f is
Y
0 (f ) := (⇠i ⇠j )2 . (3.1)
1i<jn
Note that such a field F1 always exists, for example an algebraic clo-
sure of F will do, and that, e.g. by Theorem 3.3 below, 0 (f ) 2 F and
does not depend on the choice of F1 . (This also follows from the fact that
0 (f ) is a symmetric polynomial in ⇠1 , . . . , ⇠n , and thus by a well-known
fact a polynomial in the elementary symmetric polynomials k (⇠1 , . . . , ⇠n ) =
( 1)k an k /an , k = 1, . . . , n.)
Note further that 0 (cf ) = 0 (f ) for any constant c 6= 0.
However, while the definition of 0 is simple and natural, 0 is particu-
larly useful for monic polynomials. In general, it is often more convenient
to use the following version, which by Theorem 3.5 below is a polynomial in
RESULTANT AND DISCRIMINANT OF POLYNOMIALS 11
and thus
n
Y1
n(n 1)/2 n
0 (f ) = ( 1) n an (n 1)
f (⌘j ). (3.8)
j=1
Theorem 3.11 shows that if we, more generally, for f of degree n define
n
Y1
(n)⇤
(f ) := ai 2 · (n)
(f ), (3.14)
i=0
4. Examples of discriminants
Example 4.1 (n = 1). If f = ax + b, then trivially (f ) = (f0 ) = 1.
RESULTANT AND DISCRIMINANT OF POLYNOMIALS 15
a b c
(f ) = a 1
R(f, f 0 ) = a 1
2a b 0 = b2 4ac (4.1)
0 2a b
and
b2 4ac ⇣ b ⌘2 c
2
0 (f ) =a (f ) = = 4 . (4.2)
a2 a a
Note that the standard formula for finding the roots of ax2 + bx + c = 0 can
be written
p
b 1p b± (f )
x± = ± 0 (f ) = .
2a 2 2a
Example 4.3 (n = 3). If f (x) = ax3 + bx2 + cx + d, then Theorem 3.3
yields
a b c d 0
0 a b c d
(f ) = a 1
R(f, f 0 ) = a 1
3a 2b c 0 0
0 3a 2b c 0
0 0 3a 2b c
= b2 c 2 4ac3 4b3 d + 18abcd 27a2 d2 . (4.3)
0 (f ) = (f ) = 4p3 27q 2 .
Example 4.7 (n = 4). If f (x) = ax4 + bx3 + cx2 + dx + e, then Theorem 3.3
yields
a b c d e 0 0
0 a b c d e 0
0 0 a b c d e
(f ) = a 1
R(f, f 0 ) = a 1
4a 3b 2c d 0 0 0
0 4a 3b 2c d 0 0
0 0 4a 3b 2c d 0
0 0 0 4a 3b 2c d
= b2 c 2 d 2 4 b2 c 3 e 4 b3 d3 + 18 b3 cde 27 b4 e2 4 ac3 d2 + 16 ac4 e
+ 18 abcd3 80 abc2 de 6 ab2 d2 e + 144 ab2 ce2 27 a2 d4
+ 144 a2 cd2 e 128 a2 c2 e2 192 a2 bde2 + 256 a3 e3 . (4.4)
Example 4.8 (n = 4). If f (x) = x4 + bx3 + cx2 + dx + e is monic, then
(4.4) simplifies slightly to
0 (f ) = (f ) = R(f, f 0 )
= b2 c 2 d 2 4 b2 c 3 e 4 b3 d3 + 18 b3 cde 27 b4 e2 4 c 3 d2
+ 16 c4 e + 18 bcd3 80 bc2 de 6 b2 d2 e + 144 b2 ce2
27 d4 + 144 cd2 e 128 c2 e2 192 bde2 + 256 e3 .
Example 4.9 (n = 4). If f (x) = x4 + px2 + qx + r is monic and without
third degree term, then (4.4) simplifies further to
0 (f ) = (f ) = 4 p3 q 2 27 q 4 + 16 p4 r + 144 pq 2 r 128 p2 r2 + 256 r3 .
Example 4.10. Let f (x) = xn + px + q for some n 2. Then f 0 (x) =
nxn 1 + p and, using Theorem 1.14 with h(x) = x/n, Theorem 1.13 and
Example 1.16, at least if F has characteristic 0,
0
( 1)n(n 1)/2
(f ) = Rn,n 1 (f, f ) = Rn,n 1 (x
n
+ px + q, nxn 1
+ p)
n 1
= Rn,n 1 p(1 1/n)x + q, nx +p
1)2
= ( 1)(n nn 1
R1,n 1 p(1 1/n)x + q, nxn 1 + p
⇣ ⌘
n 1
= ( 1)(n 1) n
n 1
n( q)n 1 + p p(1 1/n)
= nn q n 1
+ ( 1)n 1
(n 1)n 1 n
p .
Since the right hand side is a polynomial in p and q with integer coefficient,
the final formula holds for all fields and all n 2. Consequently,
0 (f ) = (f ) = ( 1)(n 1)(n 2)/2
(n 1)n 1 n
p + ( 1)n(n 1)/2 n n 1
n q .
Note the special cases in Examples 4.2, 4.5 and 4.9 (with p = 0). The next
case is the quintic in Bring’s form:
5
0 (x + px + q) = (x5 + px + q) = 44 p5 + 55 q 4 .
RESULTANT AND DISCRIMINANT OF POLYNOMIALS 17
Example 4.11. It follows from Remark 3.13 and Example 4.7 that
⇤
= a0 2 4 a2 a1 2 a0 1 4 a3 a2 2 a1 a0 2 + 18 a3 a2 1 a1 1 a0 1 27 a23 a2 2 a1 2
4 a4 a3 2 a2 a0 2 + 16 a4 a3 2 a22 a1 2 a0 1 + 18 a4 a3 1 a2 1 a1 a0 2
80 a4 a3 1 a1 1 a0 1 6 a4 a2 2 a0 1 + 144 a4 a2 1 a1 2
27 a24 a3 2 a2 2 a21 a0 2 + 144 a24 a3 2 a2 1 a0 1 128 a24 a3 2 a1 2
192 a24 a3 1 a2 2 a1 1 + 256 a34 a3 2 a2 2 a1 2 a0 + . . . , (4.5)
where the omitted terms have at least one factor ak with k 5. In particular,
if f (x) = a4 x4 + a3 x3 + a2 x2 + a1 x + a0 has degree at most 4, this formula
without “. . . ” is an exact formula for ⇤ (f ). Setting a4 = 0 we find that if
deg(f ) 3, then
⇤
(f ) = a0 2 4 a2 a1 2 a0 1 4 a3 a2 2 a1 a0 2 + 18 a3 a2 1 a1 1 a0 1 27 a23 a2 2 a1 2 ,
which is equivalent to (4.3). Similarly, setting also a3 = 0, if deg(f ) 2,
then
⇤
(f ) = a0 2 4 a2 a1 2 a0 1 ,
which is equivalent to (4.1).
Further, each (fi ) = 1, while (gj ) < 0 by Definition 3.2, since gj has two
roots ⇠ and ⇠ with (⇠ ⇠)2 < 0. ⇤
For example, this leads to the following classifications for low degrees. (In
these examples, “complex” means “non-real”.)
Example 5.2 (n = 2). For a real quadratic polynomial f ,
• (f ) > 0 () 0 (f ) > 0 () f has two distinct real roots;
• (f ) < 0 () 0 (f ) < 0 () f has no real root and two conju-
gate complex roots.
18 SVANTE JANSON
Appendix A
Alternative proof of Theorem 3.11. Assume first an 1 6= 0; thus f (x) =
an 1 xn 1 + · · · + a0 has degree n 1.
Let f" (x) = ( "x+1)f (x) for an indeterminate ". Trivially, ( "x+1) =
1, and Example 1.16 shows
1 ⇤
R( "x + 1, f ) = ( ")n 1
f (1/") = ( 1)n f (")
with f ⇤ defined by (1.9) with n replaced by n 1. Hence, Theorem 3.10
yields
(n)
(f" ) = (f" ) = (f )(f ⇤ ("))2 = (n 1) (f )f ⇤ (")2 .
Both sides are polynomials in a0 , . . . , an 1 and ", so we may here put " = 0
and obtain (n) (f ) = (n 1) (f )f ⇤ (0)2 . Since f ⇤ (0) = an 1 , this proves the
result when an 1 6= 0. The general case follows, because both sides of (3.12)
are polynomials in a0 , . . . , an 1 . ⇤