Li Fenglin
School of Mathematical Sciences, University of Science and Technology of China, Hefei 230026, China
Abstract: The deletion-restriction method was used to classify hyperplane arrangements with the top degree Betti number of its complements being small.
Keywords: hyperplane arrangement; deletion-restriction method; Betti number
Given a hyperplane arrangement A={H0,…,Hd} inn, letM(A)=n∪Hibe the complement of A andbn(A)=bn(M) be then-th Betti number ofM.Dimca and Papadima classified the hyperplane arrangements in the casesbn(A)=1 or 2[1]. The purpose of this paper is to extend their results to the casesbn(A)=3,4 or 5 using the deletion-restriction method.
We use [x0,…,xn] to denote the coordinates inn.For each hyperplaneHi, letlidenote its reduced defining equation.
Theorem 1.1Let A be a hyperplane arrangement innandbn(A) be itsn-th Betti number of its complement.
①bn(A)=3 if and only if it is one of the following cases, up to a change of coordinates and reordering of the hyperplanes:
(ⅰ)li(x)=xifor 0≤i≤n,ln+1(x)=x0+x1+x2.
(ⅱ)li(x)=xifor 0≤i≤n,ln+1(x)=x0+x1,ln+2(x)=a0x0+x1witha0≠0,1.
②bn(A)=4 if and only if it is one of the following cases, up to a change of coordinates and reordering of the hyperplanes:
(ⅰ)li(x)=xifor 0≤i≤n,ln+1(x)=x0+x1+x2+x3.
(ⅱ)li(x)=xifor 0≤i≤n,ln+1(x)=x0+x1,ln+2(x)=x0+x2.
(ⅲ)li(x)=xifor 0≤i≤n,ln+1(x)=x0+x1,ln+2(x)=x2+x3.
(ⅳ)li(x)=xifor 0≤i≤n,ln+1(x)=x0+x1,ln+2(x)=a0x0+x1andln+3(x)=b0x0+x1witha0,b0≠0,1 anda0≠b0.
③bn(A)=5 if and only if it is one of the following cases, up to a change of coordinates and reordering of the hyperplanes:
(ⅰ)li(x)=xifor 0≤i≤n,ln+1(x)=x0+x1+x2+x3+x4.
(ⅱ)li(x)=xifor 0≤i≤n,ln+1(x)=x0+x1+x2,ln+2(x)=a0x0+x1+x2witha0≠0,1.
(ⅲ)li(x)=xifor 0≤i≤n,
ln+1(x)=x0+x1,ln+2(x)=a0x0+x1,
ln+3(x)=b0x0+x1,ln+4(x)=c0x0+x1.
Herea0,b0,c0≠0,1 anda0,b0andc0are all different numbers.
We introduce some definitions and notations as in Ref.[2]. For A a non-empty hyperplane arrangement inn, we fixH0∈A as the hyperplane at infinity. Then we can define an affine arrangement (A,H0)a, where the total space of (A,H0)aisn=nH0and the hyperplanes of (A,H0)aare {Hi∩(nH0) |Hi≠H0}.Note that the hyperplane arrangements A and (A,H0)ahave the same complement space. It is more convenient to use the affine arrangement (A,H0)ato compute the Betti numberbn(A).In the rest of the paper, we always assume the infinity hyperplaneH0is defined byx0=0 and abuse our language to identify the hyperplane arrangements A with (A,H0)a.When we say two hyperplanes are parallel to each another, we mean that they are parallel in the affine spacen=nH0.
For any hyperplane in A, sayH1, we can use it to define two hyperplane arangements A′:=A
Dimca and Papadima classified hyperplane arrangements A whenbn(A)=1 or 2. We recall the related results here.
Lemma 2.1[1, Corollory 4]With the above notations and assumptions, we have the following:
①bn(A)=bn(A′)+bn-1(A″).
②bn(A)>0 if and only if A is essential.
③ Ifbn(A)>0, thend≤n+bn(A)-1.
Lemma 2.2[2, Proposition 1.1]If A is an arrangement in2, thenwherenkdenotes the number ofk-fold intersection points.
Theorem 2.1[1, Corollory 4]Let A be a hyperplane arrangement inn.
①bn(A)=1 if and only ifd=nand up to a linear coordinate change we haveli(x)=xifor all 0≤i≤n.
②bn(A)=2 if and only ifd=n+1 and up to a linear coordinate change and reordering of the hyperplanes we haveli(x)=xifor 0≤i≤nandln+1(x)=x0+x1.
By Lemma 2.1③, we know thatd≤n+bn(A)-1.On the other hand,bn(A)>0 implies rank(A)=n, sod≥n.To classify the hyperplane arrangements whenbn(A)=3,4 or 5, we analysis all the hyperplane arrangements withn≤d≤n+4.
In this subsection, we compute the top degree Betti number of hyperplane arrangement complement ford=n,n+1,n+2.
① Ifd=nand rank(A)=n, thenbn(A)=1.Sincebn(A)=1 impliesd≤n, the converse statement is also true. Hence up to a coordinate change, there is only one case whenbn(A)=1:
d=nandli(x)=xifor all 0≤i≤n.
② Whend=n+1, after a change of coordinate we may assumeli(x)=xifor 0≤i≤nandln+1(x)=a0x0+…+anxn.
Theorem 3.1Let A be a hyperplane arrangement as above, thenbn(A)=#{ai|ai≠0}.
ProofAfter a coordinate change, we may writeln+1(x)=a0x0+…+amxmwithai≠0 for 0≤i≤mandai=0 fori>m.It is easy to seeM(A)=M(Am)×(*)n-m, where
Am={li(x)=xifor 0≤i≤m,
lm+1(x)=a0x0+…+amxm}.
Hencebn(A)=bm(Am) by Künneth formula, so we only need to computebm(Am).Using hyperplane {xm=0} to produce a triple of arrangement, we getbm(Am)=1+bm-1(Am-1), where
Am-1={li(x)=xifor 0≤i≤m-1,
lm(x)=a0x0+…+am-1xm-1}.
Clearly Am-1has the same type of Am, so we can continue the induction. Note thatb2(A2)=3 by Lemma 2.2, hencebm(Am)=m+1.
③ For a hyperplane arrangement withd=n+2 andbn(A)>0, up to a change of coordinates we assume
A={li(x)=xifor 0≤i≤n,
ln+1(x)=a0x0+…+anxn,
ln+2(x)=b0x0+…+bnxn}.

r1r2…rvstw
where allkiare different nonzero numbers andr1≥r2≥…≥rv.Setu=r1+r2+…+rv.
Theorem 3.2For the above hyperplane arrangement, we have


ProofWe first introduce some notations. Form≤n, set
Am={li(x)=xifor 1≤i≤m,
lm+1(x)=a0+a1x1+…+amxm,
lm+2(x)=b0+b1x1+…+bmxm},
wherea0,…,an,b0,…,bnare exactly the fixed elements in the above matrix.
For hyperplane arrangement Am, we always use hyperplane {xm=0} to produce triple of arrangement. For Am′, ifbm≠0, then we can find a coordinate changeBmsuch thatx0,…,xm-1being fixed andlm+2(x) becomingxm.Ifam≠0, letAmdenote the similar coordinate change forln+1(x).
SinceM(A)=M(Au+s+t-1)×(*)w, by Künneth formula we getbn(A)=bu+s+t-1(Au+s+t-1).Hence we can eliminate the lastwcoordinates.
For A′u+s+t-1, using coordinate changeBu+s+t-1does not change the equation oflu+s+t(x).Note thatlu+s+t(x) hass+ucoefficients being non-zero. Then by Theorem 3.1 we getbu+s+t-1(A′u+s+t-1)=s+u.Since A″u+s+t-1=Au+s+t-2,by induction we have
bn(A)=(s+u)t+bs+u-1(As+u-1).
Using a similar argument forswe get
bn(A)=st+tu+su+bu-1(Au-1).
For A′u-1, after usingAu-1the new equation oflu+1(x) is
(k1-kv)+(k1-kv)x1+…+
(kv-1-kv)xr1+…+rv-1-1+kvxu-1.
By Theorem 3.1 we knowbu-1(A′u-1)=r1+…+rv-1+1.We can repeat this argument and use induction to computebu-2(A″u-1). Note that when we eliminate the lastr2+…+rvcoordinates, {1+x1+…+xr1-1=0} and {k1+k1x1+…+k1xr1-1=0} give the same hyperplane. By Theorem 3.1, the corresponding (r1-1)-th Betti number isr1.Hence we have
bu-1(Au-1)=(r1+…+rv-1+1)rv+…+
(r1+1)r2+r1=
This completes our proof.
In this subsection, we use Theorems 3.1 and 3.2 to prove Theorem 1.1. In the rest of paper, we always assumeli(x)=xifor 0≤i≤n.
Whend=norn+1, the conclusion is clear.
Whend=n+2 we use Theorem 3.2 to find out all hyperplane arrangements withbn(A)≤5.We use the same notations as in Theorem 3.2. Note thatu≥v.Ifv≥3, thenbn(A)≥6.So we only need to analysis hyperplane arrangements withv≤2.
① The casev=2.
Ifr1≥2, then there is only one case satisfyingbn(A)≤5:
(ⅰ)bn(A)=5,ln+1(x)=x0+x1+x2andln+2(x)=a0x0+x1+x2,a0≠0,1.
Ifr1=r2=1 ands+t≥2, thenbn(A)≥7.Hence there are only two cases under this condition:
(ⅰ′)bn(A)=5,ln+1(x)=x0+x1+x2andln+2(x)=a0x0+x1,a0≠0,1, which is equivalent to the case (ⅰ) after a coordinate change.
(ⅱ)bn(A)=3,ln+1(x)=x0+x1andln+2(x)=a0x0+x1,a0≠0,1.
② The casev=1.
Ifu≥3, thenbn(A)≥6.Ifu=2 ands+t≥2, then we also havebn(A)≥6.
Since hyperplanesln+1(x) andln+2(x) are different, under the conditionsv=1 andu=2 there is only one case:
(ⅲ)bn(A)=4,ln+1(x)=x0+x1+x2andln+2(x)=x0+x1.
Since hyperplanesln+1(x) andln+2(x) are not coordinate hyperplanes, then we haves≥1 andt≥1 ifu=1.Ifs≥2 ort≥2, thenbn(A)≥6.Hence we must haves=1=t:
(ⅲ′)bn(A)=4,ln+1(x)=x0+x1andln+2(x)=x0+x2.It easy to see after a change of coordinate this is equivalent to (ⅲ).
③ The casev=0.
Whenv=0 we must haves=2=t:
(ⅳ)bn(A)=4,ln+1(x)=x0+x1andln+2(x)=x2+x3.
Note that in Case (ⅲ) there exists a point which has exactlynhyperplanes passing through it, but in this case there is no such points. Hence Cases (ⅲ) and (ⅳ) are not projective equivalent.
We now analyze hyperplane arrangements withd=n+3.To havebn(A)≤5, the deleted arrangement A′ with respect toln+3(x) is a hyperplane arrangement withn+2 hyperplanes. Note thatbn(A′)<5 by Lemma 2.1①. Hence there are only three possible cases for A′ by the above discussion. For Case (iii) or (iv), we can use the hyperplaneln+3(x) to produce a triple of arrangement. The coordinate hyperplanes restricting toln+3(x) contributes at leastn-1 hyperplanes. If this number isn, then by Theorem 3.1 we havebn(A)≥6.If this number isn-1, thenln+3(x) must have the formaxi+xj.But in this caseln+1(x) andln+2(x) contributes at least another hyperplane for the restricted arrangement, so we always havebn(A)≥6.For Case (ii), ifln+3(x) is not parallel to {x1=0}, then up to a coordinate change we can reduce it to the above discussion. Hence the only possibility is Case (ii) withln+3(x) being parallel to {x1=0}:
(ⅴ)bn(A)=4 andd=n+3,ln+1(x)=x0+x1,ln+2(x)=a0x0+x1,ln+3(x)=b0x0+x1,a0,b0≠0,1,a0≠b0.
For the hyperplane arrangements withd=n+4 andbn(A)≤5, using a similar argument as above discussion we get that there is only one possible case:
(ⅵ)bn(A)=5 andd=n+4,
ln+1(x)=x0+x1,ln+2(x)=a0x0+x1,
ln+3(x)=b0x0+x1,ln+4(x)=c0x0+x1.
Herea0,b0,c0≠0,1 anda0,b0andc0are all different numbers.
Putting all these results together, Theorem 1.1 follows.
Conflictofinterest
The author declares no conflict of interest.
Authorinformation
LiFenglin(corresponding author) is a PhD candidate at Department of Mathematics, University of Science and Technology of China. His research area is singularity theory and algebraic geometry.