Bol. Soc. Mat. Mex.

(2014) 20:559–578
DOI 10.1007/s40590-014-0023-9

ORIGINAL ARTICLE

The Alexander polynomials of immersed concordant

links

Akio Kawauchi

Received: 2 April 2013 / Revised: 3 February 2014 / Accepted: 13 March 2014 /

Published online: 14 June 2014
© Sociedad Matemática Mexicana 2014

Abstract For two links bounded by an immersed concordance of annuli, we relate the
number of the double points of the immersed annuli to the beta-distance and the torsion
multi-variable Alexander polynomials of the links. This result unifies T. Kanenobu’s
announced result on the Alexander polynomial of a link with unlinking number one
and the author’s result on the Alexander polynomials of concordant links.

Keywords Link · Immersed concordance · Concordance · 4-dimensional clasp

distance · Gordian distance · Beta-distance · Torsion Alexander polynomial

Mathematics Subject Classification (2000) 57M25 · 57Q60

1 Introduction

As a link an oriented link L in the 3-sphere S 3 with components K i (i = 1, 2, . . . , r )

for r ≥ 2 is considered. A link L is immersed concordant to a link L  with components
K i (i = 1, 2, . . . , r ) if there is a smooth proper immersion

Dedicated to Francisco González-Acuña on his 70th birthday.

A. Kawauchi (B)
Osaka City University Advanced Mathematical Institute, 3-3-138 Sugimoto, Sumiyoshiku,
Osaka 558-8585, Japan
e-mail: kawauchi@sci.osaka-cu.ac.jp
560 A. Kawauchi

α : r S 1 × I → S 3 × I, I = [0, 1]

such that α(r S 1 × 0) = L × 0, α(r S 1 × 1) = L  × 1, and the image α(Sα ) of the

singular set Sα of α consists of finitely many transversely intersected double points in
S 3 × (0, 1), where r S 1 denotes the disjoint union of r copies of S 1 . It is understood
that the pairs (S 3 × 0, L × 0) and (S 3 × 1, L  × 1) are identified with the orientation
reversing pair (−S 3 , −L) and the oriented pair (S 3 , L  ), respectively. The image A
of α consisting of r immersed annuli is called an immersed concordance from L to
L  . The links L and L  are said to be concordant if α is an embedding.1 It can be seen
that for any links L and L  with the same number r of components, there is always
an immersed concordance from L to L  . Let A = ∪ri=1 Ai where Ai is the immersed
annulus in A connecting K i to K i by re-indexing the components K i (i = 1, 2, . . . , r )
of L  . The double point number c(A) of A is defined to be the cardinality of α(Sα ).
The 4-dimensional clasp distance c4 (L , L  ) between L and L  is defined by

c4 (L , L  ) = min{c(A)| A is an immersed concordance from L to L  .}

In particular, the 4-dimensional clasp number c4 (L) of L is the 4-dimensional clasp

distance c4 (L , L  ) with L  a trivial link. By definition of the 4-dimensional clasp
number, the inequalities

|c4 (L) − c4 (L  )| ≤ c4 (L , L  ) ≤ c4 (L) + c4 (L  )

are established for all links L , L  with the same number of components. It is obvious
that L and L  are concordant if and only if c4 (L , L  ) = 0. The Gordian distance
u(L , L  ) between L and L  is the minimal number of crossing changes needed to
obtain L  from L. In particular, the unlinking number u(L) of L is the Gordian distance
u(L , L  ) with L  a trivial link. By definition of the unlinking number, the inequalities

|u(L) − u(L  )| ≤ u(L , L  ) ≤ u(L) + u(L  )

are established for all links L , L  with the same number of components. If a link L  is
obtained from a link L by n crossing changes, then we have an immersed concordance
A from L to L  with c(A) = n by considering the trace of the crossing changes from
L to L  in S 3 × I . Thus, we have

u(L , L  ) ≥ c4 (L , L  ).

Let E = E(L) = cl(S 3 \N (L)) be the compact exterior of L for a tubular neigh-
borhood N (L) of L in S 3 . The first homology H1 (E) is a free abelian group of
rank r and has a basis given by choices of oriented meridians of L. We explain
here the graded multi-variable Alexander polynomials of L ([6, Chapter 7]). Let
 = Z [t1±1 , t2±1 , . . . , tr±1 ] be the integral group ring Z [H1 (E)] where ti represents a
meridian of K i . The link module of L is the -module H1 ( Ẽ) for the universal free

1 In an earlier work, “concordant” is called “cobordant”.

The Alexander polynomials 561

abelian covering Ẽ → E, namely the regular covering associated with the Hurewicz
epimorphism π1 (E) → H1 (E). By the Noetharian property of , the link module
HL = H1 ( Ẽ) is finitely generated over . For a finitely generated -module H ,
let β(H ) denote the -rank of H , namely the Q()-dimension of the Q()-vector
space H  Q() for the quotient field Q() of . Let T H be the -torsion part
of H , which is also finitely generated over  by the Noetharian property of . Let
D H be the -submodule of T H consisting of all elements x such that f i x = 0 for
coprime elements f i ∈  (i = 1, 2, . . . , s) for some s ≥ 2. Let B H = H/T H
be the -torsion-free part of H . For any non-negative integer d, the d-th character-
istic polynomial (d) (H ) is defined in [6] using a -presentation matrix of H . The
zeroth polynomial (0) (H ) is simply denoted by (H ). It is standard to use the nota-
.
tion f = f  for elements f, f  ∈  which are equal up to the units ±t1n 1 t2n 2 . . . trnr
(n 1 , n 2 , . . . , nr ∈ Z ) of . Let T (H ) = (T H ). For convenience, we list some
known facts on properties of the graded characteristic polynomials which are often
used in this paper.
Facts on properties of the graded characteristic polynomials
(1) For every short -exact sequence 0 → T  → T → T  → 0 of finitely generated
.
torsion -modules T  , T, T  , we have (T ) = (T  )(T  ).
(2) For every finitely generated -module H , we have (d) (H ) = 0 for all d < β(H )
.
and (d) (T H ) = (d+β(H )) (H ) for all d ≥ 0.
.
(3) For every finitely generated -module H with H = D H , we have (H ) = ±1.
(4) Let ( X̃ , X̃  ) be a regular Z r -covering of a compact polyhedral pair (X, X  ). If
Hd (X, X  ) = 0, then the -module Td = Hd ( X̃ , X̃  ) for  = Z [Z r ] is a finitely
generated torsion -module with (Td )(1, 1, . . . , 1) = ±1.
Explanation on how facts (1)–(4) are obtained. For the proof of (1), see [5, Lemma
2.4]. For the proof of (2), apply [6, 7.2.7 (3)] to the natural short exact sequence
0 → T H → H → B H → 0. The proof of (3) is given by the induction on the
minimal number m(H ) of -generators of H . In fact, if m(H ) = 1, then there are
coprime elements f i ∈  (i = 1, 2, . . . , s) with s ≥ 2 and a -generator x of H
such that f i x = 0 for all i, so that there is a -epimorphism / f i  → H . By
Fact (1), (H ) is a factor of the element f i for every i. Using that the elements f i
.
(i = 1, 2, . . . , s) are coprime, we see that (H ) = ±1. If m(H ) ≥ 2, then we choose
a -submodule H ⊂ H with e(H ) = 1 and m(H/H  ) = m(H )−1. By induction on
 
. . . .
m(H ), we have (H  ) = (H/H  ) = ±1, so that (H ) = (H  )(H/H  ) = ±1
by Fact (1) as desired. The proof of (4) is given in [5, Lemma 2.1] (see also [2] for a
similar argument). This concludes the explanation.
.
The identity T (H ) = (β(H )) (H ) given by Fact (3) is always a non-zero element
of  and useful in our argument. Let β(L) = β(HL ) which is called the β-rank of L.
(d) (d)
Then 0 ≤ β(L) ≤ r − 1. The d-th Alexander polynomial  L =  L (t1 , t2 , . . . , tr )
of L is just the d-th characteristic polynomial (d) (HL ). Then (d) L = 0 for all d with
(β(L))
0 ≤ d ≤ β(L) and the β(L)-th Alexander polynomial  L is equal to the zeroth
Alexander polynomial T (HL ), which we call the torsion Alexander polynomial of
L and denoted by TL . The zeroth Alexander polynomial  L is usually called the
Alexander polynomial of L. By definition,  L = 0 if and only if β(L) = 0. The
boundary polynomial b(L) ∈  of L is the zeroth characteristic polynomial of the
562 A. Kawauchi

torsion -module H1 (∂ Ẽ). The peripheral polynomial p(L) ∈  of L is the zeroth

characteristic polynomial of the image im H1 (∂ Ẽ) of the natural -homomorphism
H1 (∂ Ẽ) → H1 ( Ẽ) which is a torsion -module. The peripheral polynomial p(L) of L
is a factor of the boundary polynomial b(L) as well as a factor of the torsion Alexander
polynomial TL , and will be shown in Lemma 2.1 (4) to be a concordance invariant.
Let A be an immersed concordance from L to L  . For every i (i = 1, 2, . . . , r ), let

ci = ci (A) = c(Ai , A j ),
1≤ j≤r, j =i

where c(Ai , A j ) denotes the number of the intersection double points of the immersed
annuli Ai and A j . We note that

ci − |Link(K i , K j ) − Link(K i , K j )|
1≤ j≤r, j =i

is a non-negative even integer by the well-known properties on the intersection number

and the linking number. For an i, the pair (K i , L\K i ) is algebraically split if the
linking number Link(K i , K j ) = 0 for all j = i, and the link L is algebraically
split if (K i , L\K i ) is algebraically split for every i. For example, if (K i , L\K i ) is
algebraically non-split and (K i , L  \K i ) is algebraically split, then we have ci ≥ 1.
Let
⎧
⎪
⎪ ci if neither (K i , L\K i ) nor (K i , L  \K i )
⎪
⎪
⎪
⎪ or is algebraically split,
⎨
ci − 1 if either, but not both, of (K i , L\K i )
δi =
⎪
⎪ or (K i , L  \K i ) is algebraically split,
⎪
⎪
⎪
⎪ max{0, ci − 2} if both (K i , L\K i ) and (K i , L  \K i )
⎩
are algebraically split.

The peripheral polynomial of the immersed concordance A is the element


r
p(A) = (ti − 1)δi ∈ .
i=1

For any non-zero element f ∈  we can write it as


r
f = f (t1 , t2 , . . . , tr ) = (ti − 1)n i g
i=1

for some integers n i ≥ 0 (i = 1, 2, . . . , r ) and an element g ∈  which does not have

any element ti − 1 (i = 1, 2, . . . , r ) as a factor. Then the elementary factor e( f ) of
f is defined by e( f ) = ri=1 (ti − 1)εi where εi ∈ {0, 1} is the modulo 2 reduction
of n i . If f = 0, then we take e( f ) = 0. The peripheral polynomial p(L , L  ) of the
link pair (L , L  ) is defined to be the elementary factor p(L , L  ) = e( p(A)) of the
The Alexander polynomials 563

peripheral polynomial p(A) of the immersed concordance A, which will be shown

later to be independent of a choice of A and determined only by the link pair (L , L  ).
For a Laurent polynomial f = f (t1 , t2 , . . . , tr ) ∈ , we denote

f ∗ = f (t1−1 , t2−1 , . . . , tr−1 ) ∈ .

.
It is a classical result due to Blanchfield [1] that (d)
L = L
(d)∗
for all d, so that in
. ∗ . ∗
particular,  L =  L and  L = ( L ) . In [5], the author showed the following result
T T

in a slightly different form, whose difference is explained soon after Lemma 2.1 (see
also Theorem 5.4.2 in [13]).
Concordance Theorem If L and L  are concordant, then we have
.
TL TL  = f f ∗

for an element f ∈ .
T. Kanenobu announced the following result in [4].
Unlinking Number One Theorem. For an r -component link L with u(L) = 1,
we have  L = 0 for r > 2, and for r = 2 there is an element f ∈  such that

∗
. ff if Link(L) = ±1
L =
(t1 − 1)(t2 − 1) f f ∗ if Link(L) = 0.

The gap factor of a link pair (L , L  ) is the elementary factor g(L , L  ) = e(TL TL  )
of the product TL TL  . The β-distance of a link pair (L , L  ) is the number

β(L , L  ) = |β(L) − β(L  )| ≤ r − 1.

The following theorem is our main theorem.

Theorem 1.1 We have

u(L , L  ) ≥ c4 (L , L  ) ≥ β(L , L  ) ≥ 0.

Further, if c4 (L , L  ) = β(L , L  ), then we have

.
TL TL  = g(L , L  ) f f ∗ and g(L , L  ) = e( p(L) p(L , L  ) p(L  ))

for an element f ∈ .
It is noted that another estimate of the Gordian distance u(L , L  ) from below is
given by different invariants of a link in [7]. As a corollary to Theorem 1.1, we obtain
the following result.
4-dimensional Clasp Number One Theorem For an r -component link L with
c4 (L) = 1, we have  L = 0 for r > 2, and for r = 2 there is an element f ∈  such
that

∗
. ff if Link(L) = ±1
L =
(t1 − 1)(t2 − 1) f f ∗ if Link(L) = 0.
564 A. Kawauchi

We note that if r = 2, c4 (L) = 1 (or u(L) = 1) and Link(L) = 0, then we have

Link(L) = ±1 because of the following inequalities

u(L) ≥ c4 (L) ≥ |Link(L)|.

Then Unlinking Number One Theorem follows directly from 4-dimensional Clasp
Number One Theorem because if  L = 0 and u(L) = 1, then r = 2 and u(L) =
c4 (L) = β(L , O) = β(O) = 1.
In Sect. 2, some computations of the boundary polynomial and the peripheral poly-
nomial are done and it is explained how Concordance Theorem and 4-dimensional
Clasp Number One Theorem are derived from Theorem 1.1. Throughout Sect. 3, the
proof of Theorem 1.1 is done. In Sect. 4, we show three corollaries to Theorem 1.1 and
examples on the unlinking number, the 4-dimensional clasp number and the peripheral
polynomials.

2 Computing the boundary polynomial and the peripheral polynomials

The following lemma is used to compute the boundary polynomial and the peripheral
polynomial of a link.
Lemma 2.1 We have the following (1)–(3).

(1) The boundary polynomial b(L) of a link L is given by b(L) = ri=1 (ti − 1)εi
where εi ∈ {0, 1} (i = 1, 2, . . . , r ), and εi = 1 if and only if the pair (K i , L\K i )
is algebraically split.
(2) If β(L) = 0, then p(L) = b(L).
(3) If β(L) = r −1, then L is algebraically split, b(L) = ri=1 (ti −1) and p(L) = 1.
(4) Let f L ∈  be the factor of TL obtained from TL by removing all the non-unit
prime factors g1 ∈  of TL with g1 (1, 1, . . . , 1) = 1. Then p(L) is a factor of
. .
f L . If L is concordant to a link L  , then we have f L = f L  and p(L) = p(L  ).
In (4), we take an element f˜L ∈  with TL = f L f˜L . Then f˜L (1, 1, . . . , 1) = ±1.
.
Counting the identity f L = f L  given by (4) and also known by [5], we obtain the iden-
. .
tities f˜L = f˜L  hh ∗ and TL = TL  hh ∗ for an element h ∈  with h(1, 1, . . . , 1) = ±1
from Concordance Theorem. This last identity is an exact form of Concordance The-
orem given in [5].
Proof To see (1), we note that the lift of the torus component Ti of ∂ E around K i
to ∂ Ẽ consists of components homeomorphic to S 1 × R or R 2 according s to whether
(K i , L\K i ) is algebraically split or not. Hence we have H1 (∂ Ẽ) ∼= k=1 /(tik − 1)
for the members i k (k = 1, 2, . . . , s) in {1, 2, . . . , r } with εik = 1 and b(L) =
s
k=1 (ti k − 1). To see (2), since

β(L) = 0, H0 ( Ẽ) = D H0 ( Ẽ) ∼

= /(t1 − 1, t2 − 1, . . . , tr − 1), D H1 (∂ Ẽ) = 0,

Blanchfield duality [1] implies that there is a (t1 , t2 , . . . , tr )-anti-monomorphism

B H2 ( Ẽ, ∂ Ẽ) → hom(B H1 ( Ẽ), ) = 0

The Alexander polynomials 565

and there is a (t1 , t2 , . . . , tr )-anti-isomorphism

T H2 ( Ẽ, ∂ Ẽ)/D H2 ( Ẽ, ∂ Ẽ) ∼

= hom(T H0 ( Ẽ)/D H0 ( Ẽ), Q()/) = 0.

Thus, we see that H2 ( Ẽ, ∂ Ẽ) = D H2 ( Ẽ, ∂ Ẽ). Then, because D H1 (∂ Ẽ) = 0, the
boundary map ∂∗ : H2 ( Ẽ, ∂ Ẽ) → H1 (∂ Ẽ) is the zero map. Hence the natural homo-
morphism H1 (∂ Ẽ) → H1 ( Ẽ) is injective, so that we have p(L) = b(L). To see (3),
let (1) = Z [t, t −1 ] be the t-variable Laurent polynomial ring and Ẽ (1) → E the
infinite cyclic covering associated with the epimorphism H1 (E) → Z sending every
oriented meridian of L in ∂ E to 1 ∈ Z . Then, the (1) -rank β(1) (L) of the (1) -
module H1 ( Ẽ (1) ) is r − 1 (cf. [6, 7.3.12]), which implies that the torsion Alexander
polynomial (1) of the (1) -module H1 ( Ẽ (1) ) has (1) (1) = ±1. Thus, the natural
homomorphism

H1 (∂ Ẽ (1) ) = ((1) /(t − 1))r → H1 ( Ẽ (1) )

must be trivial, meaning that the longitude of K i in ∂ E is the boundary of a 2-chain

in E and hence Link(K i , L\K i ) = 0, for every i. Thus, L is algebraically split. By
(1), we have b(L) = ri=1 (ti − 1). To see (4), we note that the peripheral poly-
nomial p(L) is a factor of TL which does not have any non-unit prime element
g1 ∈  with g1 (1, 1, . . . , 1) = 1. Hence p(L) is a factor of f L . To see that p(L) is
a concordance invariant, let E(A) = cl(S 3 × I \N (A)) be the exterior of a concor-
dance A from L to L  where N (A) denotes a tubular neighborhood of A in S 3 × I .
Let RA = cl(∂ E(A)\(E(L) × 0 ∪ E(L  ) × 1)) which is homeomorphic to both
(∂ E(L)) × I and (∂ E(L  )) × I . The universal free abelian coverings Ẽ(L) → E(L)
and Ẽ(L  ) → E(L  ) extend to the universal free abelian covering Ẽ(A) → E(A).
Since the natural homomorphism H1 (∂ Ẽ(L)) → H1 ( R̃A ) is a -isomorphism, the
natural homomorphism H1 ( Ẽ(L)) → H1 ( Ẽ(A)) sends the image im H1 (∂ Ẽ(L)) of
the natural homomorphism H1 (∂ Ẽ(L)) → H1 ( Ẽ(L)) onto the image im H1 ( R̃A ) of
the natural homomorphism H1 ( R̃A ) → H1 ( Ẽ(A)). Let A T be the zeroth polynomial

of T H1 ( Ẽ(A)), and p(A) the zeroth polynomial of im H1 ( R̃A ). Let f A ∈  be the

factor of A T obtained from T by removing all the non-unit prime factors g ∈ 
A 1
of AT with g (1, 1, . . . , 1) = 1. We have
1

H∗ (E(A), E(L) × 0) = H∗ (E(A), E(L  ) × 1) = 0.

By Fact (4), the -module T∗ = H∗ ( Ẽ(A), Ẽ(L) × 0) is a torsion -module with

(T∗ )(1, 1, . . . , 1) = ±1. In the homology exact sequence of the pair ( Ẽ(A), Ẽ(L)×
0), let T2 be the image of the connecting homomorphism

∂∗ : T2 = H2 ( Ẽ(A), Ẽ(L) × 0) → T H1 ( Ẽ(L) × 0),

and T1 the image of

j∗ : T H1 ( Ẽ(A) × 0) → H1 ( Ẽ(A), Ẽ(L) × 0) = T1 .

566 A. Kawauchi

By Fact (1), (Td )(1, 1, . . . , 1) = ±1 (d = 1, 2). From the short -exact sequence

0 → T H1 ( Ẽ(L) × 0)/T2 → T H1 (E(A)) → T1 → 0

.
and Fact (1), the identity f L = f A is obtained. The zeroth polynomial of the -
module T2 = T2 ∩ im H1 (∂ Ẽ(L)) is a unit of , because by Fact (1) the zeroth
polynomial (T2 ) is a common factor of the zeroth polynomials (T2 ) and p(L)
which is coprime. Since the -epimorphism im H1 (∂ Ẽ(L)) → im H1 ( R̃A ) induces a
-isomorphism

im H1 (∂ Ẽ(L))/T2 ∼
= im H1 ( R̃A ),
. . .
we have p(L) = p(A) by Fact (1). Similarly, we have f L  = f A and p(L  ) = p(A).
. .
Thus, we have f L = f L  and p(L) = p(L  ). This completes the proof. 

The following lemma shows that the peripheral polynomial of a link pair is calcu-
lable by the linking numbers modulo 2 of the immersed concordant links and hence
is independent of a choice of an immersed concordance.

Lemma 2.2 The peripheral polynomial p(L , L  ) of a pair (L , L  ) has the form


r
p(L , L  ) = (ti − 1)εi ,
i=1

where εi ∈ {0, 1} is determined by

⎧   
⎪ Link(K i , L\K i ) + Link(K
⎪
⎪ i , L \K i ) (mod 2) if neither (K i , L\K i )
⎪
⎨   
nor (K i , L \K i ) is algebraically split,
εi = Link(K i , L\K i ) + Link(K i , L  \K i ) − 1 (mod 2) if either, but not
⎪
⎪
⎪
⎪ both, of (K i , L\K i ) or (K i , L  \K i ) is algebraically split,
⎩
0 if both (K i , L\K i ) and (K i , L  \K i ) are algebraically split.

Proof As observed in the introduction, we have

ci ≡ Link(K i , L\K i ) + Link(K i , L  \K i ) (mod 2).

In particular, if both (K i , L\K i ) and (K i , L  \K i ) are algebraically split, then ci is

even and max{0, ci − 2} ≡ 0 (mod 2). This completes the proof.


The following corollary is useful in deriving 4-dimensional Clasp Number One

Theorem from Theorem 1.1.

Corollary 2.3 Assume that c4 (L , L  ) = β(L , L  ) = r − 1 with β(L) = 0 and

β(L  ) = r − 1. Then we have the following (1) and (2).

(1) If L is algebraically split, then we have g(L , L  ) = ri=1 (ti − 1).
(2) If Link(K i , L\K i ) is odd for every i, then we have g(L , L  ) = 1.
The Alexander polynomials 567

Proof By the assumption of (1), we have p(L) = ri=1 (ti − 1) and p(L  ) = 1 by
(1), (2) and (3) of Lemma 2.1, and p(L , L  ) = 1 by Lemma 2.2, obtaining (1) by
Theorem 1.1. For the assumption of (2), we have p(L) = p(L  ) = 1 by (1), (2) and
(3) of Lemma 2.1, and p(L , L  ) = 1 by Lemma 2.2, obtaining (2) by Theorem 1.1.


We are in a position to explain how Concordance Theorem and 4-dimensional Clasp
Number One Theorem are derived from Theorem 1.1.
Deriving Concordance Theorem and 4-dimensional Clasp Number One Theorem
from Theorem 1.1. To derive Concordance Theorem from Theorem 1.1, assume that
L and L  areconcordant. Then we have c4 (L , L  ) = β(L , L  ) = 0 and p(L , L  ) = 1.
.
Since p(L) is a concordance invariant, we have p(L) = p(L  ), so that g(L , L  ) = 1.
Thus, Concordance Theorem is obtained from Theorem 1.1. To derive 4-dimensional
Clasp Number One Theorem from Theorem 1.1, let L be an r -component link with
c4 (L) = 1. By Theorem 1.1, we have

1 ≥ β(L , O) ≥ β(O) − β(L) = r − 1 − β(L)

for an r -component trivial link O. Hence if r > 2, then β(L) ≥ r − 2 > 0 and
 L = 0. For r = 2, assume  L = 0. Then β(L) = 0 and c4 (L) = β(L , O) = 1.
Since T (O) = 1 and |Link(L)| ≤ c4 (L) = 1, the desired result follows directly
from Theorem 1.1 and Corollary 2.3.

3 Proof of Theorem 1.1

Throughout this section, the proof of Theorem 1.1 will be done. Let A be an immersed
concordance from L to L  with n = c(A) = c4 (L , L  ). We assume that

β(L  ) = β(L) + b

for b = β(L , L  ) = |β(L) − β(L  )|. Since every transversely intersected double point
in A is topologically represented by the cone vertex of a Hopf link, we slide the double
points in A into L  × 1. Then, we obtain from Ai a connected oriented proper planar
surface Fi in S 3 × I for every i such that Fi ∩ F j = ∅ for every i = j and the boundary
∂ F of the surface F = ∪ri=1 Fi is given by ∂ F = (−L) × 0 ∪ L  × 1, where L  is the
link which is regarded as a connected sum L  = L  #sj=1 S H u H
j #k=1 Uk for the link L


and the Hopf links S j ( j = 1, 2, . . . , s) and Uk (k = 1, 2, . . . , u) with the following

H H

conditions (1)–(3) (see Fig. 1 for an illustration of an immersed concordance A and

Fig. 2 for an illustration of the surface F):
(1) s + u = n.
(2) The connected sum connects one component of L  and one component of S H j for
 H
every j and one component of L and one component of Uk for every k.
(3) The components of every Hopf link S Hj except the arc used for the connected sum
belong to the same component of F and the components of every Hopf link UkH
except the arc used for the connected sum belong to distinct components of F.
568 A. Kawauchi

Fig. 1 An immersed
concordance A

Fig. 2 The surface F

We attach a (±1)-framed 2-handle h 2j to S 3 × 1 to make a (±1)-twist on S H j

producing a two-component trivial link. Then the surgery of S 3 × 1 on the 2-handles
h 2j ( j = 1, 2, . . . , s) change the link L  into the split link of the links (L  #k=1
u U H)×1
k
and an s-component trivial link O × 1. We assume that the attaching core circle
s

of the 2-handle h 2j has linking number 0 with S H j . By adding s mutually disjoint

disks in S × 1 bounded by O × 1 to the surfaces Fi (i = 1, 2, . . . , r ), we obtain
3 s

planar surfaces Fi (i = 1, 2, . . . , r ) in a 4-manifold X = S 3 × I ∪sj=1 h 2j with

boundary ∂ X = S 3 × 0 ∪ S 3 × 1 such that the surface F  = ∪ri=1 Fi has boundary
∂ F  = (−L) × 0 ∪ (L  #k=1 u U H ) × 1. Further, we add the connected sum bands
k
Bk (k = 1, 2, . . . , u) used for the connected sum (L  #k=1 u U H ) × 1 in S 3 × 1 to the
k
  
surface F . The resulting surface F = F ∪k=1 Bk in X has r components such that the
u

boundary ∂ F  is given by ∂ F  = (−L)×0∪ L + ×1, where L + ×1, is a link in S 3 ×1,

split into the link L  ×1 and the Hopf links UkH ×1 (k = 1, 2, . . . , u). After deforming
The Alexander polynomials 569

F  into a proper surface in X , let Y = cl(X \N (F  )) be the compact 4-manifold for a

tubular neighborhood N (F  ) = F  ×D 2 of F  in X . For the link L, let L = L (0) ∪L (1) ,
where L (0) is the sublink of L consisting of all components K such that (K , L\K )
is algebraically split and L (1) is the sublink of L consisting of the other components.
Applying the same notation to the link L + , we have L + = (L + )(0) ∪ (L + )(1) with
(L + )(0) = (L  )(0) and (L + )(1) = (L  )(1) ∪uk=1 UkH . Let M0 be the 0-surgery manifold
of S 3 along L (0) , and

M = cl(M0 \N (L (1) ))

for a tubular neighborhood N (L (1) ) of L (1) . Similarly, let M0 be the 0-surgery manifold
of S 3 along L (0) = (L + )(0) , and

M  = cl(M0 \N ((L  )(1) )), 

M+ = cl(M0 \N ((L + )(1) )).

Then we have a connected sum decomposition M+  = M  # u E(U H ) for the Hopf

k=1 k
link exteriors E(UkH ) (k = 1, 2, . . . , u). Since ∂Y is a torus sum of link exteriors
E(L) and E(L + ), and the product F  × S 1 , we construct a 4-manifold

W = M × [−1, 0] ∪ Y ∪ M+ × [1, 2]

pasting M × [−1, 0] and Y along E(L) × 0 and pasting Y and M+  × [1, 2] along

E(L + ) × 1. Let G be a possibly disconnected proper surface in W obtained from a

push off of F  by attaching disks to the components of L (0) and (L + )(0) = (L  )(0) .
See Fig. 3 for an illustration of the surface G.
Then the boundary ∂ W of the 4-manifold W is a torus sum of M × (−1), M+  ×2

and the product G × S . If G has a 2-sphere component, namely if there is an S × S 1

1 2

component in G × S 1 , then we paste the 4-manifold B 3 × S 1 with B 3 the 3-ball to

Fig. 3 The surface G

570 A. Kawauchi

it and hence assume that G × S 1 has no S 2 × S 1 components. For simplicity, we

take M = M × (−1), P = G × S 1 , and M+  = M  × 2. By construction, the
+
maximal free abelian covering Ẽ(L) over E(L) extends to a free abelian covering
 ) over (W ; M, P, M  ). For a compact submanifold pair (W  , W  ) of
(W̃ ; M̃, P̃, M̃+ +
W , the homology H∗ (W̃  , W̃  ) of the lift (W̃  , W̃  ) to W̃ forms a finitely generated
-module. Let β∗ (W̃  , W̃  ) denote the -rank of H∗ (W̃  , W̃  ). At first, we note that
H1 ( P̃) is a finitely generated torsion -module, which is seen from the following
lemma, because every component P1 of P satisfies the assumption of the lemma after
a suitable permutation of the indexes of t j ( j = 1, 2, . . . , r ).

Lemma 3.1 Let G 1 be a 2-sphere with n(≥ 2) holes, and P1 = G 1 × S 1 . Let the
fundamental group π1 (P1 ) have a presentation with generators xi (i = 1, 2, . . . , n)
and relators r1 = x1 x1−1 and ri = x1 xi x1−1 xi−1 (i = 2, . . . , n) where x1 and xi (i =
2, 3, . . . , n) are represented by a loop p × S 1 ( p ∈ G 1 ) and loops in G 1 × 1 (1 ∈ S 1 ),
respectively. Let P̃1 → P1 be the covering associated with a homomorphism γ :
π1 (P1 ) → Z r such that γ∗ (x1 ) = t1 and γ∗ (xi ) = u i is a monomial with coefficient
+1 in t j ( j = 1, 2, . . . , r ) such that u = t1k for any integer k. Then we have the
.
Alexander polynomial (H1 ( P̃1 )) = (t1 − 1)n−2 . In particular, H1 ( P̃1 ) is a finitely
generated torsion -module.

Proof
 We use the Fox free calculus [3] (see also [6, 7.1.5]). The Jacobian (n, n)-matrix
γ
∂ri /∂ x j ∗ with entries in  is given by
⎛ ⎞
0 0 0 ... 0
⎜ 1 − u 2 t1 − 1 0 ... 0 ⎟
⎜ ⎟
⎜ 1 − u3 − ... 0 ⎟
⎜ 0 t 1 1 ⎟.
⎝ ... ... ... ... ... ⎠
1 − un 0 0 . . . t1 − 1

This matrix is a presentation matrix of a -module M admitting a short exact sequence

0 → H1 ( P̃) → M → ε() → 0,

where ε() = (t1 − 1, t2 − 1, . . . , tr − 1) denotes the fundamental ideal which is

the torsion-free -module of rank one obtained as the kernel of the homomorphism
.
 → Z sending every ti to 1. By Fact (2), the Alexander polynomial (H1 ( P̃1 )) =
(1)
 (M), which is obtained from the second elementary ideal of the Jacobian matrix
by taking the smallest principal ideal. The second elementary ideal is

((t1 − 1)n−1 , (1 − u i )(t − 1)n−2 (i = 1, 2, 3, . . . , n)).

.
Hence we have (H1 ( P̃)) = (t1 − 1)n−2 , completing the proof of Lemma 3.1.
We need the following computations of -ranks, where the identity s + u = n is
noted.


Lemma 3.2 (1) β1 (∂ W̃ ) = 2β(L) + b + u.

The Alexander polynomials 571

(2) β1 (W̃ ) ≤ β(L).

(3) β1 (∂ W̃ ) ≤ β2 (W̃ , ∂ W̃ ) + β1 (W̃ ).
(4) β2 (W̃ , ∂ W̃ ) = n + u + β1 (W̃ ).

Proof Since every solid torus S 1 × D 2 attaching to E(L) or E(L + ) used to construct
M or M+  lifts to the disjoint union of copies of R 1 × D 2 in M̃ or M̃  , we have
+
β(L) = β1 ( M̃) and

β(L + ) = β1 ( M̃+

) = β1 ( M̃  ) + u = β(L  ) + u = β(L) + b + u.

 and P = G × S 1 and H ( P̃) is a finitely generated

Since ∂ W is a torus sum of M, M+ 1
torsion -module, we have

β1 (∂ W̃ ) = β1 ( M̃) + β1 ( M̃+ ) = 2β(L) + b + u,

showing (1). Since H1 (W, M) = 0, we see from Fact (4) that β1 (W̃ , M̃) = 0, which
implies that β(L) = β1 ( M̃) ≥ β1 (W̃ ), showing (2). Using the exact sequence

H2 (W̃ , ∂ W̃ ) → H1 (∂ W̃ ) → H1 (W̃ ),

we have (3). To see (4), we note by Blanchfield duality [1] that β2 (W̃ , ∂ W̃ ) = β2 (W̃ )
and β3 (W̃ ) = β1 (W̃ , ∂ W̃ ). We see that β1 (W̃ , ∂ W̃ ) = 0 by the exact sequence

H1 (W̃ , M̃) → H1 (W̃ , ∂ W̃ ) → H0 (∂ W̃ , M̃)

of the triad (W̃ , ∂ W̃ , M̃) and β1 (W̃ , M̃) = β0 (∂ W̃ , M̃) = 0. The Euler characteristic
χ (W ) of W is given by χ (W ) = n + u. In fact, W is a union of M × [−1, 0], Y ,
M+  × [1, 2] and some copies of B 3 × S 1 pasting along E(L) × 0, E(L  ) × 1 and
+
some copies of S 2 × S 1 whose Euler characteristics are 0 except Y . Hence we have
χ (W ) = χ (Y ) = χ (X ) − χ (F  ), for Y = cl(X \N (F  )). Since X is obtained from
S 3 × I by attaching the 2-handles h 2j ( j = 1, 2, . . . , s), we have χ (X ) = s. Because
the surface F  is obtained from F by attaching s disks along the boundaries and by
spanning u bands where F is homeomorphic to r annuli with n open disks removed,
we have χ (F  ) = −n + s + u − 2u = −2u. Thus, we have χ (W ) = s + 2u = n + u
as desired. Then since βd (W̃ ) = 0 (d = 1, 2), it follows that

β2 (W̃ ) − β1 (W̃ ) = n + u,

showing (4). This completes the proof. 

By Lemma 3.2, we have the following inequalities:

2β(L) + b + u = β1 (∂ W̃ ) (by (1))

≤ β2 (W̃ , ∂ W̃ ) + β1 (W̃ )
= 2β1 (W̃ ) + n + u (by (3) and (4))
≤ 2β(L) + n + u (by (2)).
572 A. Kawauchi

This means the inequality β(L , L  ) = b ≤ n = c4 (L , L  ), giving the first half

of Theorem 1.1. Further, if this inequality is replaced by the equality, namely, if
β(L , L  ) = b = n = c4 (L , L  ), then we have the identity:

(∗) β2 (W̃ , ∂ W̃ ) + β1 (W̃ ) = β1 (∂ W̃ ).

We need the following lemma:

Lemma 3.3 (Exactness Lemma) Under the identity (*), the natural exact sequence

∂∗ i∗
H2 (W̃ , ∂ W̃ ) −→ H1 (∂ W̃ ) −→ H1 (W̃ )

induces an exact sequence

∂∗ i ∗
T H2 (W̃ , ∂ W̃ ) −→ T H1 (∂ W̃ ) −→ T H1 (W̃ ).

This lemma is shown by the observation that the identity (*) implies that the -
homomorphism

B H2 (W̃ , ∂ W̃ ) → B H1 (∂ W̃ )

induced from ∂∗ is a monomorphism, where we note that the -module B H induced

from every -module H is a torsion-free -module by definition. Let H = H1 (∂ W̃ ),
and T  the image of ∂∗ : T H2 (W̃ , ∂ W̃ ) → T H1 (∂ W̃ ) = T H . By [5, Theorem 3.1],
we have
.
T (H ) = (T H ) = (T  )(T  )∗ .

To calculate T (H ), we need the following lemma.

. .
Lemma 3.4 (1) TL = p(L)T (H1 ( M̃)), TL  = p(L  )T (H1 ( M̃  )).
.
(2) T (H ) = p(A)T (H1 ( M̃))T (H1 ( M̃  )).
.
(3) (H1 ( P̃)) = p(A).

Proof To see (1), let V be the solid tori used to construct M from E(L). Since H1 (Ṽ ) =
0, the homology exact sequences of the pairs ( Ẽ(L), ∂ Ẽ(L)), ( M̃, Ṽ ) connected by the
excision isomorphisms H1 ( Ẽ(L), ∂ Ẽ(L)) ∼ = H1 ( M̃, Ṽ ) induce the following short
exact sequence

0 → im H1 (∂ Ẽ(L)) → H1 ( Ẽ(L)) → H1 ( M̃) → 0.

Similarly, we have the following short exact sequence.

0 → im H1 (∂ Ẽ(L  )) → H1 ( Ẽ(L  )) → H1 ( M̃  ) → 0.
The Alexander polynomials 573

Applying Fact (1) to the short exact sequences restricted to the torsion parts of these
short exact sequences, we have (1). To see (2), let S = ∂ M ∪ ∂ M+  which is a union

of tori. Then we have H1 ( S̃) = 0. By excision, we have

 
H1 (∂ W̃ , S̃) ∼
= H1 ( M̃, ∂ M̃) H1 ( P̃, S̃) 
H1 ( M̃+ 
, ∂ M̃+ ).

Since there is a natural exact sequence

H1 ( S̃) = 0 → H1 (∂ W̃ ) → H1 (∂ W̃ , S̃) → H0 ( S̃)

and H0 ( S̃) = D H0 ( S̃), we see from Facts (1) and (3) that
.
T (H ) = T (H1 (∂ W̃ )) = T (H1 (∂ W̃ , S̃)).

Similarly, we have
.
T (H1 ( M̃, ∂ M̃)) = T (H1 ( M̃)),
.
 , ∂ M̃  )) = .
 )) =
T (H1 ( M̃+ + T (H1 ( M̃+ T (H1 ( M̃  )).

By Lemma 3.1, note that H1 ( P̃, S̃) = T H1 ( P̃, S̃) and H1 ( P̃) = T H1 ( P̃). By the
.
exact sequence of the pair ( P̃, S̃), we have (H1 ( P̃, S̃)) = (H1 ( P̃)). Thus, we
have
.
T (H ) = (H1 ( P̃))T (H1 ( M̃))T (H1 ( M̃  )).

Assuming (3), we complete the proof of (2). To see (3), we note that the component G i
of G obtained from Ai is a 2-sphere with ci +2 holes if (K i , L\K i ) nor (K i , L  \K i ) is
algebraically split; a 2-sphere with ci + 1 holes if either, but not both, of (K i , L\K i ) or
(K i , L  \K i ) is algebraically split; and a 2-sphere with ci holes if both (K i , L\K i ) and
(K i , L  \K i ) are algebraically split. In the last case, recall that ci is always even and
we omitted the case ci = 0. Then, (3) is confirmed from Lemma 3.1. This completes
the proof of Lemma 3.4. 

By Lemmas 3.3 and 3.4, we have
p(A)2 TL TL 
.
= p(A)2 p(L) p(L  )T (H1 ( M̃))T (H1 ( M̃  ))
.
= p(A) p(L) p(L  )T (H )
.
= p(A) p(L) p(L  )(T  )(T  )∗
.
= e( p(L) p(L , L  ) p(L  ))gg ∗

for some non-zero element g ∈ , so that

.
T (L)T (L  ) = g(L , L  ) f f ∗ ,
g(L , L  ) = e( p(L) p(A) p(L  )) = e( p(L) p(L , L  ) p(L  ))

for some f ∈ . This completes the proof of Theorem 1.1.

574 A. Kawauchi

4 Corollaries to Theorem 1.1 and related examples

We consider a pair (L , L  ) of an r -component algebraically split link L with  L = 0

and an r -component completely split link L  with components K i (i = 1, 2, . . . , r )
such that c4 (L , L  ) ≤ r −1. Since β(L , L  ) = β(L  ) = r −1, we see from Theorem 1.1
that c4 (L , L  ) = β(L , L  ) = β(L  ) = r − 1. For every r ≥ 2, there are lots of
such pairs (L , L  ). For example, from any given knot K and any r ≥ 2, Shimizu
constructed in [12] an algebraically split link L of r components, called a lassoed
link associated with K , such that  L = 0 and u(L , L  ) = r − 1 for the completely
splittable link L  consisting of K and the (r − 1)-component trivial link, where we
have u(L , L  ) = c4 (L , L  ) = β(L , L  ) = r − 1 by Theorem 1.1. In the following
corollary, we will concern with a relationship between the Alexander polynomials of
such a link pair (L , L  ).
Corollary 4.1 Let (L , L  ) be a pair of an r -component algebraically split link L
with  L = 0 and an r -component completely split link L  with components K i (i =
1, 2, . . . , r ) such that c4 (L , L  ) ≤ r − 1. Let δ K i be the product of all mutually
.
distinct prime factors a of  K i such that a ∗ = a and the exponent of a in the prime
decomposition of  K i is odd. Then we have c4 (L , L  ) = β(L , L  ) = β(L  ) = r − 1
and

. 
r 
r
L = (ti − 1) δ K j f f ∗
i=1 j=1

for an element f ∈ , where the Alexander polynomial  K i is regarded as an integral

Laurent polynomial in ti for every i.
Proof By Theorem 1.1, we have the identities c4 (L , L  ) = β(L , L  ) = β(L  ) = r −1
.
and hence  L TL  = g(L , L  )gg ∗ for an element g ∈ . By Lemmas 2.1 and 2.2,
p(L) = i=1 (ti − 1), p(L  ) = 1 and p(L , L  ) = 1, so that g(L , L  ) = p(L) =

r
r ∗ . . ∗
i=1 (ti − 1). Using that  K  =  K i , we see that  K i = δ K i di di for an element
i
di ∈  and hence

. 
r
.
TL  =  K j = δ L  dd ∗
j=1

for δ L  = rj=1 δ K j and d = rj=1 d j . If aa ∗ divides bb∗ for non-zero elements
a, b ∈ , then the quotient bb∗ /aa ∗ is written as the form cc∗ for an element
c ∈ , which can be easily shown by taking the prime decompositions of a and
b. Since TL  (1, 1, . . . , 1) = ±1, we see that TL  is coprime with g(L , L  ) and
hence TL  divides gg ∗ . Let hh ∗ be the quotient gg ∗ /dd ∗ for an element h ∈ .
.
Then  L δ L  = g(L , L  )hh ∗ . For every prime factor a of δ L  , the product a 2 divides
. .
hh ∗ because a divides hh ∗ and satisfies a ∗ = a. Hence δ 2L  = δ L  δ ∗L  divides hh ∗ .
The quotient is denoted by the product f f ∗ for an element f ∈ . Then we have
.
 L = g(L , L  )δ L  f f ∗ , completing the proof. 

The Alexander polynomials 575

For a link L with r components, the fraction

ˆ L (t) =  L (t)/(t − 1)r −1 =  L (t, t, . . . , t)/(t − 1)r −2



is known to be an integral polynomial, called the Hosokawa polynomial of L (see [6]).

ˆ L (t) is always even. In fact, the
We know that the multiplicity of the factor t − 1 in 
statement that σ1 (L) ≡ κ̂1 (L) (mod 2) in Lemma 5.7 of [8] implies this assertion.
Then the following corollary is obtained from Theorem 1.1.

Corollary 4.2 If the one-variable Alexander polynomial  L (t) of a link L with r (≥ 2)

components is a non-zero polynomial, then

u(L) ≥ c4 (L) ≥ β(L , O) = r − 1.

ˆ L (t) has the form

Further if c4 (L) = r − 1, then the Hosokawa polynomial 

ˆ L (t) =. f (t) f (t −1 )


for an integral polynomial f (t) in t up to multiplications of ±t i (i ∈ Z ).

Recall that the group order of the first homology H1 (M L ) of the double branched
covering space M L of S 3 branched along L coincides with the absolute value | L (−1)|
by taking the group order of an infinite abelian group to be 0 (see [6]).

Corollary 4.3 If the first homology H1 (M L ) of the double branched covering space
M L of S 3 branched along a link L with r (≥2) components is a finite abelian group,
then

u(L) ≥ c4 (L) ≥ β(L , O) = r − 1.

Further if c4 (L) = r − 1, then the group order |H1 (M L )| has the form

|H1 (M L )| = 2r −1 n 2

for an integer n.

In the case of a link L with r = 2 and u(L) = 1 (implying c4 (L) = 1), the
latter half of Corollary 4.3 has been observed by Kohn [9]. The following example
concerns a computation on the unlinking number, the 4-dimensional clasp number and
the peripheral polynomials for the links illustrated in Fig. 4 together with the notation
of “linkinfo”2 in the bracket, whose Alexander polynomials are given in [11].

Example 4.4 (1) Let L = 632 (L6a4) (the Borromean rings), and O the 3-component
trivial link. Since u(L) ≤ 2, β(L) = 0 and β(O) = 2, we have u(L) = c4 (L) =
β(L , O) = 2 by the first half of Theorem 1.1. We have p(O) = 1. Since L is

2 http://www.indiana.edu/~linkinfo/.
576 A. Kawauchi

Fig. 4 Links used in Example 4.4

algebraically split, we see from Lemma 2.2 that p(L , O) = 1. The latter half
assertion of Theorem 1.1 is confirmed as follows:
.
 L = g(L , O) = (t1 − 1)(t2 − 1)(t3 − 1) = p(L).

We can also see u(L) = c4 (L) = 2 from the inequality u(L) ≤ 2 and the following
fact that the link L is algebraically split, but not link-homotopically trivial (see
[10]).
(2) Let L = 721 (L7a6), and O the 2-component trivial link. Then we have

.
 L = 1 − t1 − t2 + (1 − t1 − t2 )t1 t2 + (t1 t2 )2

which cannot be written as g(L , O) f f ∗ . Hence by Theorem 1.1 we obtain

u(L) ≥ c4 (L) > β(L , O) = 1.

Since u(L) ≤ 2, we have u(L) = c4 (L , O) = 2.

(3) Let L = 722 (L7a5), and O the 2-component trivial link. Then we have

.
 L = 1 − t1 − t2 + (3 − t1 − t2 )t1 t2 + (t1 t2 )2 .

Since u(L) ≤ 1, β(L) = 0 and β(O) = 1, we see from the first half of Theo-
rem 1.1 that u(L) = c4 (L) = β(L , O) = 1. The latter half assertion of Theo-
rem 1.1 is confirmed as follows:
.
 L = (t1 + t2−1 − 1)(t1−1 + t2 − 1), g(L , O) = p(L) = p(O) = p(L , O) = 1.
The Alexander polynomials 577

(4) Let L = 835 (L8a16), and O the 3-component trivial link. Then we have
u(L , L 0 ) = 1 for the split link L 0 consisting of the unoriented (2, 4)-torus link
T2,4 and a trivial knot. Note that β(L) = 0, β(L 0 ) = 1 and
.
TL 0 = t1 t2 + 1,  L = (t1 − 1)(t2 − 1)(t3 − 1)(t1 t2 + 1),

where t1 and t2 are represented by the meridians of the sublink T2,4 of L 0 . Then
we have u(L , L 0 ) = c4 (L , L 0 ) = β(L , L 0 ) = 1 by the first half of Theorem 1.1
and the latter half assertion of Theorem 1.1 is confirmed as follows:
.
 L TL 0 = g(L , L 0 )(t1 t2 + 1)(t1−1 t2−1 + 1)

where g(L , L 0 ) = e( p(L) p(L , L 0 ) p(L 0 )) = (t1 − 1)(t2 − 1)(t3 − 1). Since
the sublink T2,4 of L 0 has the linking number ±2 and the corresponding sublink
L T in L has the linking number ±1, and Link(K , L T ) = 0 for the component
K = L\L T , we see from Lemma 2.2 that p(L , L 0 ) = (t1 − 1)(t2 − 1). Using
that b(L 0 ) = t3 − 1 and the longitude of O bounds a disk in the exterior E(L 0 ),
we see that p(L 0 ) = 1, so that p(L) = t3 − 1. From the linking number of T2,4 ,
we obtain that u(L 0 ) = 2 and hence u(L) ≤ 3. On the other hand, we have
β(L , O) = β(O) = 2. Examining the form of  L , we see from Theorem 1.1
that c4 (L) > β(L , O) = 2. Thus, we have that u(L) = c4 (L) = 3.
(5) Let L = 939 (L9a54), and O the 3-component trivial link. Then we have
.
 L = (t1 − 1)(t2 − 1)(t3 − 1)(t12 − t1 + 1),

which cannot be written as g(L , O) f f ∗ . Hence by Theorem 1.1 we have

u(L) ≥ c4 (L) > β(L , O) = 2.

Since u(L) ≤ 3, we have u(L) = c4 (L) = 3.

(6) L = 9312 (L9a53) which is an algebraically split link, and O the 3-component
trivial link. Since u(L) ≤ 2 and β(L , O) = 2, we have u(L) = c4 (L) =
β(L , O) = 2 and
.
 L = g(L , O)(t1 − 1)(t1−1 − 1)

where g(L , O) = (t1 − 1)(t2 − 1)(t3 − 1) = p(L) and p(L , O) = p(O) = 1.

(7) Let L = 9321 (L9n27) which is an algebraically split link, and O the 3-component
.
trivial link. Then we have  L = 0, u(L) ≤ 1, β(O) = 2, β(L) = 1, T (L) =
t1 − 1 and p(O) = 1. By the first half of Theorem 1.1, we have u(L) = c4 (L) =
β(L , O) = 1. The latter half of Theorem 1.1 is confirmed as follows:
.
TL = g(L , O) = e( p(L) p(L , O) p(O)) = t1 − 1.

By Lemma 2.2, p(L , O) = 1, so that p(L) = t1 − 1.

578 A. Kawauchi

Acknowledgments This work was supported by Grant-in-Aid for Scientific Research(A) (24244005).
The author would like to thank the referees for suggesting lots of devices to understand this paper easier.

References

1. Blanchfield, R.C.: Intersection theory of manifolds with operators with applications to knot theory.
Ann. Math. 65, 340–356 (1957)
2. Cochran, T.D., Orr, K.E., Teichner, P.: Knot concordance, Whitney towers and L 2 -signatures. Ann.
Math. 157, 433–519 (2003)
3. Crowell, R.H., Fox, R.H.: Introduction to knot theory, Ginn and Co. (1963); Reissue, Graduate Texts
in Mathematics, vol. 57, Springer, New York (1977)
4. Kanenobu, T.: The Alexander polynomials of links with unlinking number one (in Japanese), Abstract
in the fall term meeting of the Mathematical Society of Japan, 10–11 (1993)
5. Kawauchi, A.: On the Alexander polynomials of concordant links. Osaka J. Math. 15, 151–159 (1978)
6. Kawauchi, A.: A Survey of Knot Theory. Birkhäuser, Basel, Boston, Berlin (1996)
7. Kawauchi, A.: Distance between links by zero-linking twists. Kobe J. Math. 13, 183–190 (1996)
8. Kawauchi, A.: The quadratic form of a link. Proc. Low Dimens. Topol. Contemp. Math. 233, 97–116
(1999)
9. Kohn, P.: Unlinking two component links. Osaka J. Math. 30, 741–752 (1993)
10. Milnor, J.W.: Link groups. Ann. Math. 59, 177–195 (1954)
11. Rolfsen, D.: Knots and Links, (revised edition). AMS Chelsea Press (2003)
12. Shimizu, A.: The complete splitting number of a lassoed link. Topol. Appl. 159, 959–965 (2012)
13. Turaev, V.G.: Reidemeister torsion in knot theory. Russ. Math. Surv. 41, 119–182 (1986)