A polynomial invariant of a homology 3-sphere defined by Reidemeister torsion
Teruaki Kitano
Soka Univ.
26th Jan., 2017
Contents
Johnson theory
Reidemeister torsion Casson invariant
Polynomial defined by Reidemeister torsion Brieskorn homology sphere
1/n-surgery along the figure-eight knot SL(2;C)-Casson invariant
Witten’s topolgical field theory
Reidemeister torsion
1930’s Reidemeister, Franz, de Rham: Classification of lens space
1961-2 Milnor:
Hauptvermutung for polyhedrons
Alexander polynomial=Reidemeister torsion 1980’s asymptotic behavior of quantum invariants 1990’s twisted Alexander polynomial=Reidemeister torsion
1996 Meng-Taubes: SW=Milnor torsion
Casson invariant and Reidemeister torsion
1985 Casson, an integral lift of Rochlin invariant Late 1980’s D. Johnson, A geometric form of Casson’s invariant and its connection to Reidemeister torsion, unpublished lecture notes (hand written).
A generalization of the determinant
Roughly speaking,
Reidemeister torsion= ”det”(an acyclic based chain complex).
acyclic: a linear isomorphism
based: a matrix is given under a basis
determinant for a linear isomorphism
V: l-dimensional vector space over a field F(=C,R).
f :V →V: a linear isomorphism How to define det(f) ?
e1, . . . ,el: a basis of V
F ∈M(l;F): the matrix defined by f undere1, . . . ,el Fore1, . . . ,el, we take their image f(e1), . . . ,f(el).
det(f,(e1, . . . ,el)) = det(F)
= det (f(e1), . . . ,f(el)/e1, . . . ,el).
Under fixing a basis ofV, e1, . . . ,el, we have an identification
∧lV ∼=F.
f(e1)∧· · ·∧f(el) = det (f(e1), . . . ,f(el)/e1, . . . ,el)e1∧· · ·∧el
determinant for a short exact sequence
0→V1 →f1 V2 →f2 V3 →0: an exact sequence of finite dimensional vector spaces overF.
bi ⊂Vi: a basis of Vi for i = 1,2,3 f2−1(b3)⊂V2: a lift of b2
(f1(b1),f2−1(b3)): another basis of V2
τ = det(
f1(b1),f2−1(b3)/b2) Remark
τ does not depend on a choice of a lift f2−1(b3).
Reidemeister torsion for a chain complex
F: a field
a chain complex of finite dimensional vector spaces overF C∗:
0−→∂m+1 Cm −→∂m Cm−1 ∂−→m−1. . .−→∂2 C1 −→∂1 C0 −→∂0 0 ci: a basis of Ci
Definition
A chain complexC∗ isacyclic if
Im∂i+1 =Ker∂i (Zi(C∗) =Bi(C∗),Hi(C∗) ={0})
Assumption: C∗ is acyclic.
Fix bi for Bi(C∗) =Zi(C∗).
Exact sequence
0−→Zi(C∗)−→Ci −→∂i Bi−1(C∗)−→0 ebi−1: a lift of bi−1 in Ci
(bi,ebi−1): another basis ofCi (
bi,ebi−1/ci )
: the transformation matrix from ci to (bi,ebi−1)
[bi,ebi−1/ci] = det (
bi,ebi−1/ci )
:
Torsion of a based chain complex
Definition
The torsionτ(C∗) of a based chain complex C∗ with c∗ τ(C∗) = Πi:odd[bi,ebi−1/ci]
Πi:even[bi,ebi−1/ci] ∈F\ {0}. Lemma
τ(C∗) does not depend on choices of lifts {bi}.
Lemma
τ(C∗) does not depend on choices of {bi}. Proof.
Assumeb′q is another basis ofBq.
In the definition of τ(C∗), the difference between bq and b′q is related to the followings only two parts:
[b′q,bq−1/cq]
= [bq,bq−1/cq][
b′q/bq] [bq+1,b′q/cq+1] = [bq+1,bq/cq+1][
b′q/bq] Since[
b′q/bq]
appears in the both of the denominator and the numerator of the definition, they are cancelled.
Reidemeister torsion for a CW-complexe
X:a finite(=compact) CW-complex X˜ →X:universal covering
π1(X) acts on ˜X (from the right), cellularly.
Ci( ˜X;Z):spanned by lifts of i-cells of X overZ.
C∗( ˜X;Z): a chain complex of right Z[π1(X)]-modules.
local coefficients
ρ:π1(X)→GL(V): a finite dimensional linear representation over F
V:l-dimensional vector space over F (=C2,sl(2;C),su(2))
C∗(X;Vρ) =C∗( ˜X;Z)⊗π1(X)Vρ Assume: C∗(X;Vφ) is acyclic.
Definition
τρ(X) = τ(C∗(X;Vρ))∈F\ {0}.
well-definedness
Choices of lifts of cells Orders of cells. Choice of a basis for V Proposition
τρ(X)∈F/±det(ρ(π1(X))) does not depend on choices of lifts and orders of cells.
It depends on a choice of basis e1, . . . ,el. Then we fix e1 ∧ · · · ∧el ∈∧l
V.
topological invariant
Under same assumption, we have the following.
Theorem (Kirby-Siebenmann, Chapman)
If X is a closed manifold, thenτρ(M)∈F/±det(ρ(π1(X))) is a topological invariant of M.
Remark
For any acyclic even-dimensional unimodular representation ρ:π1(X)→SL(2l;F),τρ(X)∈F\ {0} is well defined.
Remark
In the case of V =C2, we take e1 =
(1 0
) ,e2 =
(0 1
) .
In the case of V =su(2), under a fixed Killing form, we take an orthonormal basis.
If ρ is not acyclic, then we defineτρ(X) = 0.
Definition
We callρ an acyclic representation ifC∗(X;Vφ) is acyclic.
Condition for acyclicity
If C∗(X;Vφ) is acyclic, then Euler number of H∗(X;Vφ) is 0 =∑
i
(−1)idimHi(X;Vρ)
=∑
i
(−1)idimCi(X;Vρ)
=∑
i
(−1)idimCi(X;Z)·dimV
=lχ(X).
Hereχ(X) is Euler characteristic of X. Hence we have χ(X) = 0.
There does not exist an acyclic representation for X with χ(X)̸= 0.
Johnson Theory
M : homology 3-sphere λ(M)∈Z : Casson invariant
R(X) ={ρ:π1X →SU(2);irreducible representation} Rˆ(X) =R(X)/SU(2)
M =A∪Σg B : Heegard decomposition Rˆ(M) = ˆR(A)∩Rˆ(B)⊂Rˆ(Σg)
Assume:
Rˆ(M) is a finite set.
Any point of ˆR(M) isnon degenerate.
For anyρ∈ R(M), compose it with
Adj :SU(2)→Aut(su(2)), we have
Adj ◦ρ:π1(M)→Aut(su(2)).
dimRH1(Σg;su(2)Adj◦ρ) = 6g −6 dimRH1(A;su(2)Adj◦ρ) = 3g −3 dimRH1(B;su(2)Adj◦ρ) = 3g−3 H1(M;su(2)Adj◦ρ) = {0}
volume forms
Johnson constructed volume forms volΣg onH1(Σg;su(2)Adj◦ρ), volA on H1(A;su(2)Adj◦ρ), volB on H1(B;su(2)Adj◦ρ).
By the assumption, he definedt[ρ]∈R\ {0} by
volΣg =t[ρ](volA ∧volB)∈
6g∧−6
H1(Σg;su(2)Adj◦ρ).
Geometric form of Casson invariant
Theorem (Johnson)
λ(M) = ∑
[ρ]∈Rˆ(M)
sgn(t[ρ]) t[ρ] =τAdj◦ρ(M)
Remark
Adj ◦ρ :π1M →SU(2) →Aut(su(2)) τAdj◦ρ(M) =τ(C∗(M;su(2)Adj◦ρ)) Ageometric form of Casson invariant:
∑
[ρ]∈Rˆ(M)
t[ρ]
A polynomial invariant defined by Reidemeister torsion
Johnson proposed to study
a polynomial whose zeros are {t[ρ]}:
∏
[ρ]
(t −t[ρ]),
a polynomial for Reidemeister torsion τρ(M) for an SL(2;C)-representation.
In particular he studied ∏
[ρ]
(t −τρ) for 1/n-surgeried manifold along the trefoil knot.
A polynomial defined by Reidemeister torsion
M : homology 3-sphere
τρ(M)∈C : Reidemeister torsion forM with an irreducibleSL(2;C)-representation ρ
τρ′(M) = 1/τρ(M)
Assume : ˆRSL(2;C)(M) is afinite set.
By assumption, {τρ(M) | τρ(M)̸= 0} is a finite set.
Definition
σM(t)∈C[t] is a polynomial whose roots are given by the set {
τρ′(M) = τ 1
ρ(M) | τρ(M)̸= 0
} .
Assumption : ˆRSL(2;C)(M) is an algebraic variety of dimension zero.
By using resultants, it is given as a set of algebraic numbers.
Any representation ρ in ˆRSL(2;C)(M) can be realized over an algebraic extension fieldF over Q.
τρ(M) and τρ′(M) are also algebraic numbers inF. Any Galois conjugate ofρ (under the action of Gal(F/Q)) is a representation.
If the action of Gal(F/Q) gives all Galois conjugate of τρ′ by Gal(Q({τρ′})/Q), then
σM(t)∈Q[t].
It is the minimal polynomial ofτρ′ ∈Q(τρ′).
minimal polynomial
F ⊃Q is an extension ofQ. θ∈F
Definition
The minimal polynomial ofθ is the monic polynomial of least degree among all polynomials inQ[x] having θ as a root.
Remark
The minimal polynomial ofθ exists when θ is algebraic over Q, that is, f(θ) = 0 for some non-zero polynomial f(x)∈Q[x].
Brieskorn homology 3-sphere
Σ(p,q,r) : Brieskorn homology 3-sphere {z1p+z2q+z3r = 0} ∩S5 ⊂C3 =R6 p,q,r ∈Z : pairwise coprime positive integers We may assumeq and r are odd numbers.
Σ(p,q,r) can be given by Dehn surgery along (p,q)-torus knot
Σ(p, q, | pqn + 1 | )
Now consider the case of r =|pqn+ 1|. Then
Σ(p,q,|pqn+ 1|) can be obtained by 1/n-surgery along T(p,q).
In this case any conjugacy class of irreducible representations can be represented by
ρ(a,b,k) :π1(Σ(p,q,|pqn+ 1|))→SL(2;C) with
1 0<a <p,0<b <q,a ≡b mod 2,
2 0<k <|pqn+ 1|,k ≡na mod 2,
3 tr(ρ(a,b,k)(x)) = 2 cosaπp ,
4 tr(ρ(a,b,k)(y)) = 2 cosbπq ,
5 tr(ρ(a,b,k)(µ)) = 2 cos|pqn+1kπ |.
τ
ρ(Σ(p , q , | pqn + 1 | ))
Johnson computed Reidemeister torsion τρ(a,b,k)(Σ(p,q,|pqn+ 1|)) as follows.
Theorem (Johnson)
1 A representation ρ(a,b,k) is acyclic if and only if a≡b ≡1,k ≡n mod 2.
2 For an acyclic representation ρ(a,b,k), τρ(a,b,k)(Σ(p,q,|pqn+ 1|))
= 1
2 (
1−cos aπp ) (
1−cosbπq ) (
1 + cos|pqn+1pqkπ| ).
τ
ρ(Σ(p , q , r ))
In general case, we have the following.
Proposition (Johnosn, Kitano-Tran)
τρ(a,b,c)(Σ(p,q,r)) = 1
2 (
1−cosaπp ) (
1−cosbπq ) (
1−coscπr )
= 1
16 sin2 (πa
2p
) sin2
(πb 2q
)
sin2(πc
2r
)
where(0,0,0)<(a,b,c)<(p,q,r), a≡b≡c ≡1mod 2.
σ
(p,q,r)(t )
From hereσΣ(p,q,r)(t) is simply written as σ(p,q,r)(t).
σ(p,q,r)(t) = C ∏
(a,b,c)
(
t −16 sin2 (πa
2p )
sin2 (πb
2q )
sin2 (πc
2r ))
where the above product can be taken over the same condition (0,0,0)<(a,b,c)<(p,q,r), a ≡b ≡c ≡1 mod 2.
Now put
C(p,q,a,b)= 2 sin (πa
2p )
sin (πb
2q )
.
Normalized Chebyshev polynomial of first kind
Normalized Chebyshev polynomial of the first kind:
Tn(2cosθ) =2cosnθ.
Recursive definition:
T0(x) = 2, T1(x) =x.
Tn+1(x) =xTn(x)−Tn−1(x).
T0(x) = 2,T1(x) =x,T2(x) = x2−2,T3(x) = x3−3x.
Normalized Chebyshev polynomial of second kind
Normalized Chebyshev polynomial of the second kind:
Sn(2cosθ) = sin((n+ 1)θ) sinθ . Recursive definition:
S0(x) = 1, S1(x) = x.
Sn+1(x) =xSn(x)−Sn−1(x).
S0(x) = 1,S1(x) =x,S2(x) =x2−1,S3(x) =x3−2x. A Relation:
−
Polynomial of Σ(p , q , r )
Theorem (Kitano-Tran)
σ(p,q,r)(t) = ∏
(a,c)
Sq−1
( √ t C(p,r,a,c)
)
= ∏
(a,b)
Sr−1
( √ t C(p,q,a,b)
)
=
∏
(b,c)
Sp−1
( √ t C(q,r,b,c)
)
(p :odd)
∏
(b,c)
S¯2p−1
( √ t C(q,r,b,c)
)
(p :even) .
HereS¯2p−1(x) =S2p−1(x)/x .
Examples
σ(2,3,5)(t) =4t2 −6t + 1
σ(2,3,7)(t) =8t3 −20t2+ 12t −1
σ(2,3,11)(t) =32t5−144t4+ 224t3−140t2+ 30t −1
σ(2,3,13)(t) =64t6−352t5+ 720t4−672t3+ 280t2−42t + 1 σ(2,3,17)(t) =256t8−1920t7+ 5824t6−9152t5+ 7920t4
−3693t3+ 840t2−72t + 1
σ(2,3,19)(t) =512t9−4352t8+ 15360t7−29120t6+ 32032t5
−20592t4 + 7392t3−1320t2+ 90t −1 They are irreducible over Q.
Remark
Example (reducible example)
σ(2,5,9)(T) =256t8 −2688t7+ 9856t6−15840t5 +12192t4−4608t3+ 820t2−60t + 1 This is not irreducible as
σ(2,5,9)(T)
=(4t2−6t+ 1)
×(64t6−576t5+ 1584t4−1440t3+ 492t2−54t + 1)
=σ(2,3,5)(T)
×(64t6−576t5+ 1584t4−1440t3+ 492t2−54t + 1)
SL(2; C )-Casson invariant
ForM = Σ(2,3,6n+ 1)(n>0),
♯RˆSL(2;C)(Σ(2,3,6n+ 1)) =λSL(2;C)(Σ(2,3,6n+ 1))
=3n (Boden−Curtis) On the other hand,
λSU(2)(Σ(2,3,6n+ 1)) =n.
λSL(2;C)(Σ(2,3,6n+ 1))−2λSU(2)(Σ(2,3,6n+ 1)) =n What is the difference ?
♯RˆSL(2;R)(Σ(2,3,6n+ 1)) =n
Proposition (K.-Yamaguchi)
λSL(2;C)(Σ(p,q,r)) = 2λSU(2)(Σ(p,q,r))+♯RˆSL(2;R)(Σ(p,q,r)) ρ:π1(Σ(p,q,r))→SL(2;C) induces
¯
ρ:π1orbS2(p,q,r)→PSL(2;C).
πorb1 S2(p,q,r) =⟨ x,y,z | xp =yq =zr =xyz = 1 ⟩ Jenkins-Neumann:
Classification of PSL(2;R)-representations of triangle groups
AnySU(2)-representation that is conjugate to SL(2;R)-representation is an abelian representation.
”SL(2; R )-Casson invariant”
Problem
Can we define an SL(2;R)-Casson invariant ?
Dehn surgery along the figure-eight knot
K = 41
π1(S3\K) =⟨a,b|w−1a =bw−1⟩wherew =ba−1b−1a.
Mn(K): 1/n-Dehn surgery along K.
ρ:π1(Mn(K))→SL(2,C): irreducible representation.
µ=a∈π1(Mn(K)): meridian of K x =tr(ρ(µ))
Reidemeister torsion for M
n(K )
Reidemeister torsion ofMn(K) for ρ is given as follows.
Theorem (K.)
τρ(Mn(K)) =− 2(x−1) x4−9x2+ 4 Here x =tr(ρ(µ)).
Remark
The explicit value x is determined by the surgery condition. Remark that any complex value x can not be realized.
Letρ:π1(S3\K)→SL(2;C) be an irreducible representaion.
By taking a conjugation, we may assume ρ(a) =
[ s 1 0 s−1
]
, ρ(b) =
[ s 0
−u s−1 ]
.
Here (s,u)∈(C∗)2 is a solution of
ϕK(s,u) =u2−(u+ 1)(s2+s−2−3).
Compute the image of longitude λ= (ab−1a−1b)−1)(ba−1b−1a−1)−1,
ρ(λ) =
[ λ11 λ12 λ21 λ22
]
where
λ11= (−s−2−1)u3+ (s2−s−2)u+ (s−4+s2−s−2−1)u2+ 1, λ12=u(s +s−1)(s2+s−2−1−u),
λ21=−u2(s+s−1)ϕK(s,u),
λ22= (−s2−1)u3+ (s4−s2+s−2−1)u2+ (s−2−s2)u+ 1.
The conditionϕK(s,u) = 0 implies
trρ(λ) =s4+s−4−s2−s−2−2
=x4−5x2+ 2 wherex =s+s−1.
Here this representation π1(S3\K)→SL(2;C) can be extend as ρ:π1(Mn(K))→SL(2;C) if and only if
ρ(λn) = ρ(µ−1).
By using L=trρ(λ) =x4−5x2+ 2, ρ(λn) =
[ λn11 λ12Sn−1(L) 0 λn22
] .
Further the condition ρ(λn) = ρ(µ−1) is equivalent to the following three conditions of
1 λn11=s−1,
2 λn22=s,
3 λ12Sn−1(L) = −1.
By direct computation, we have
1 u2 −(u+ 1)(s2+s−2−3) = 0,
2 λn11=s−1,
3 L̸=−2 (i.e. x2 ̸= 4).
Lemma
The set of s satisfying the following
1 u2 −(u+ 1)(s2+s−2−3) = 0,
2 λn11=s−1
is coincided wth the set of solutions for
s +s−1 =Tn(s4+s−4−s2−s−2−2).
By the formula of Reideimeister torsion τρ = 2x −2
x2(x2−5), we consider the resultant
Pn(t) = Resx
(Tn(x4−5x2+ 2)−x,tx2(x2−5)−(2x −2)) .
A resultant
f(x) =anxn+an−1xn−1 +· · ·+a1x +a0
g(x) =bmxm+bm−1xm−1+· · ·+b1x +b0 Definition (the resultant off(x),g(x))
Res(f(x),g(x)) =
an an−1 · · · · · · a0
an an−1 · · · a1 a0
an · · ·
...
an · · · a1 a0
bm bm−1 · · · · · · b0
bm bm−1 · · · b1 b0
bm · · ·
...
bm · · · b1 b0
f(x,y),g(x,y):polynomials of 2-variable x,y By considering x is a variable andy is a constant, we can define the resultant Resx(f,g).
Proposition
There exist polynomials F(x,y),G(x,y)such that f(x,y)G(x,y) +g(x,y)F(x,y) = Resx(f,g).
Therefore a solution (x,y) off(x,y) =g(x,y) = 0 satisfies Res(f,g) = 0.
Polynomial for surgeried manifold along 4
1Mn(K) is a homology 3-sphere obtained by 1/n-surgery along K = 41.
Theorem (Kitano-Tran) σMn(K)(t) =
{
Pn(t)/(2t−3) n is odd, Pn(t)/(2t+ 1) n is even.
Here
Pn(t) =Resx(
Tn(x4−5x2+ 2)−x,tx2(x2−5)−(2x −2))
Examples
σ1(t) =t3−12t2+ 20t −8
σ2(t) =t7−56t6+ 660t5−3384t4+ 8720t3−11008t2 + 5376t −128
σ3(t) =t11−124t10+ 3036t9−31696t8+ 161024t7
−364128t6 + 152640t5+ 426752t4−262144t3
−142336t2 + 55296t −2048
σ4(t) =t15−224t14+ 10320t13−211776t12+ 2296400t11
−13570900t10+ 41172200t9−49672100t8−35529500t7 +156351000t6−113653000t5−58957800t4+ 115933000t3
−50004000t2 + 5898240t−32768
Open Problems
Problem
How strong is this polynomial σM(t) ?
How about for Brieskorn homology sphere ?
σ
M(t ) is a minimal polynomial?
Problem
How is it related with minimal poly ofτρ′ overQ?
Problem
If there exists an epimorphismφ:π1(M)→π1(M′), then σM(t)can be divided by σM′(t)?
Problem
Does σM(t) know♯RˆSL(2;C)(M) and SL(2;C)-Casson invariant?
Does τρ′ with the action of a Galois group know
♯RˆSL(2;C)(M) and SL(2;C)-Casson invariant?
General case
Problem
How can we treat it in the case of ♯Rˆ(M) = ∞? Only consider the componets of dimension 0? How to take a perturbation ?
How to treat splicing manifolds ?
In the original definition of the Casson invariant, we need to take a perturbation to get a transverse intersection. The problem is that a representation after a perturbation does not give a representation of π1(M).
Problem
Can we construct λSL(2;R)(M)for (some class of) homology 3-spheres satisfying
λSL(2;C)(Σ(p,q,r)) = 2λ(Σ(p,q,r)) +λSL(2;R)(Σ(p,q,r))?
Proposition
There exists a hyperbolic homology 3-sphere without an irreducible SL(2;R)-representation.
SL(2; R )-representations
Mn(K): 1/n-surgeried along the figure-eight knot Proposition
λSL(2;C)(Mn(K)) = 4n−1, λ(Mn(K)) = −n
λSL(2;C)(Mn(K))−2|λ(Mn(K))|= 2n−1 Checked by Mathematica:
Example (n=2,. . . ,6,8)
λSL(2;C)(Mn(K))−2|λ(Mn(K))|> ♯RˆSL(2;R)(Mn(K))>0 Example (n= 7,9)
TQFT
Problem
How can we construct TQFT for Johnson theory?
Witten’s theory
Description from the view point of topological quantum field theory: E. Witten, Topology-changing amplitudes in
(2 + 1)-dimensional gravity, Nuclear Phys. B 323(1989).