(c) Conjugatingtto each term ofat1′,t′∈T, cost zero relations. Then we basically estimate the cost of the following equatioin
at1′t=at1′t.
By the result of Lemma V.3.3, since Tr(t′t)⊂B(degt+degt′), then the cost is bounded by(2K)(degt+degt′).Notice deg(t′)6Q, then the total cost is at most
(mP)(2K)(degt+degt′)6(mP)(2K)(Q+degt).
Here we use the fact Tr(t)⊂Bdeg(t)since we order elements inZT degree lexicographically.
Recall thatG∞is a factor group ofH∞ asGis a factor group ofG∞. Denote the epimorphism fromH∞toG∞ induced by identity on generating set asψ, and then we have the following homomorphism chain:
H∞ ψ G∞ ϕ G.
Thus kerψ=hhR2iiH∞,ker(ϕ◦ψ) =hhR2∪R3iiH∞.They are all normal subgroups inH∞as well as submodules in hhAiiH∞. H∞contains a free module structure while each ofGandG∞contain a factor module of it. Eventually we will convert the word problem to a membership problem of a submodule inhhAiiH∞.
Note thatGis a factor group ofG∞and with the given presentation Lemma V.3.1, Lemma V.3.2, Lemma V.3.3 and Lemma V.4.1 all hold forG.
Since Dehn function is a quasi-isometric invariant then it enough for us to prove Proposition V.1.2 using the presentation (V.8).
Proof of Proposition V.1.2. We start with a wordw∈Gsuch that|w|=n,w=G1. WLOG we may assume
w=u1b1u2b2. . .usbsus+1
whereui∈F=F(T),bi∈A±1ands+∑s+1i=1|ui|=n. Letvi= (u1. . .ui)−1fori=1, . . . ,sandν=u1u2. . .us+1. Then we have
w=w1:=bv11bv22. . .bvssν.
The equality holds in the free group generated byA ∪T thus the cost of relations convertingwtow1is 0. Since s+∑s+1i=1|ui|=n, in particular, we have thats6n. Moreover|vi|=∑ij=1|uj|6nhence Tr(θ(vi))⊂Bn,i=1,2, . . . ,n.
Next sincew1=G1,π(w1) =π(vs+1) =1. By Lemma V.3.1,
ν=
s′ i=s+1
∏
bvii
wheres′−s6|ν|26n2,bi∈A±1, and Tr(θ(vi))⊂Bn,i=s+1, . . . ,s′. By Lemma V.3.1, the cost of convertingνto the right hand side is bounded by|ν|26n2.
Thus we let
w2:=
s′
∏
i=1bvii,s′6n2+n,Tr(θ(vi))⊂Bn,i=1, . . . ,s′. And the cost of convertingw2tow1is bounded byn2.
Next, note that allvi’s are words inF. With the help of Lemma V.3.3, we are able to organizevito its image in ¯F.
More precisely, we let
w3:=
s′
∏
i=1bvi¯i,s′6n2+n,kθ(¯vi)k6n.
Also followed by Lemma V.3.3,w2=Gw3. Let us estimate the cost of convertingw2tow3. To transformw2tow3, we need apply Lemma V.3.3 to eachbvii once. Since Area(bviib−¯i vi)6(2K)nwhich provided by Tr(θ(vi))⊂Bn, each transformation costs(2K)nrelations. We have in totals′6n+n2many conjugates to convert therefore the cost is bounded by(n2+n)(2K)n.
Now letw4be the ordered form ofw3, which in fact is also the ordered form ofw, i.e.
w3=Gw4:=
m
∏
i=1aiµi
whereµi are ordered under≺. By the discussion in Section V.2, we obtain the ordered form just by rearranging all conjugates ofA±1. Note that becausekθ(v¯i)k6n for alli, it cost at mostK2nrelations to commute any two consecutive conjugatesbvi¯i andbv¯jj by Lemma V.3.2. To sorts′conjugates we need commutes′2times. Therefore the number of relations need to commutew3tow4is bounded above thats′2K2n6(n2+n)2K2n.
The only thing remains is to compute the area ofw4. Recall thatw4 can be regarded as an element in a free T-module generated bya1, . . . ,am. w4=G1 implies that eitherw4=H∞ 1 or it lies in the submodule generated by
R4={f1,f2, . . . ,fl}which is the Gr¨obner basis of the submodule generated byR2∪R3. Ifw=H∞ 1 thenµi=/0 for
alli=1, . . . ,m. In this case Area(w4) =0. Thus
Area(w)6n2+ (n2+n)(2K)n+ (n2+n)2K2n.
We are done with this case.
Now let us consider the casew∈ hhR4iiH∞\ {1}. LetKbe a constant large enough to satisfy both Corollary IV.4.2 and Lemma V.3.2. As an element in theT-module, degw46n sincekθ(¯vi)k6nfor alli. Also recall that for an elementα∈ZT,|α|is defined to be thel1-norm of it regarded as a finite suppported function fromT toZ. Thus|w4| represents the number of conjugates inw4which iss′. Then by Corollary IV.4.2 we have
w4=H∞ l
∏
i=1fiαi,fi=a1µi1aµ2i2. . .aµmim∈R4,deg(fiαi)6n,
l i=1
∑
|αi|6s′Kn2k6(n2+n)Kn2k.
whereµi=∑lj=1αjµjiinZT. Note that fiαi is the product consisting of exactly|αi|many relators. In conclusion we have
Area(
l
∏
i=1fiαi)6
l
∑
i=1
|αi|6(n2+n)Kn2k.
Last, let us estimate the cost of converting∏li=1fiαi tow4. This process consists of two different steps: 1. converting all fiαi’s to their ordered form; 2. adding thelterms of orderedfiαi.
To transformfiαi to its ordered form, we write
αi=
∑
u∈suppαi
αi(u)u.
Let us denoteP=maxli=1|fi|,Q=maxli=1deg(fi). Then
fiαi=fi∑u∈suppαiαi(u)u=
∏
suppαi
fiαi(u)u=
∏
suppαi
fi,u=aµ1′i1aµ2i2′ . . .amµim′ =OF(fi), (V.9)
wherefi,uis the ordered form offiαi(u)uandu′i j=αiµi jhence OF(fi)is the ordered form offiαi. The first two equalities above hold in the free groupF(A ∪T)thus the cost is 0. In the third equality, applying Lemma V.4.1 (b) and (c), the cost of convertingfiαi(u)uto fi,uis bounded bym|fi|(2K)k(degfi+degu)+ (|αi(u)| −1)m2|fi|2K2(degfi+degu). Here we first conjugateuto fithen add|αi(u)|terms of fiu. Because degu6degαi,∑u∈suppαi|αi(u)|=|αi|,|suppαi|6|αi|.
Consequently the cost of the third equality of (V.9) is bounded by
u∈suppα
∑
i(m|fi|(2K)k(degfi+degu)+ (|αi(u)| −1)m2|fi|2K2(degfi+degu)) 6
∑
u∈suppαi
(m|fi|(2K)k(degfi+degαi)+ (|αi(u)| −1)m2|fi|2K2(degfi+degαi))
=|suppαi|m|fi|(2K)k(degfi+degαi)+
∑
u∈suppαi
(|αi(u)| −1)m2|fi|2K2(degfi+degαi)
=|suppαi|m|fi|(2K)k(degfi+degαi)+ (|αi| − |suppαi|)m2|fi|2K2(degfi+degαi) 6|αi|m|fi|(2K)k(degfi+degαi)+|αi|m2|fi|2K2(degfi+degαi)
=|αi|(m|fi|(2K)k(degfi+degαi)+m2|fi|2K2(degfi+degαi)) 6|αi|(mP(2K)kn+m2P2K2n).
The last inequality is obtained by the condition deg(fiαi)6n, i.e degfi+degαi6n.
The forth equality of (V.9) is adding allfi,u’s up. Since
degfi,u6degfiαi 6n,|fi,u|6|αi(u)||fi|6|αi||fi|6|αi|P,
by Lemma V.4.1 (a), the cost of adding|suppαi|terms of fi,uis bounded by(|suppαi| −1)m2(|αi|P)2K2n. Here we use the fact that the size of the addition of any step is bounded by|fiαi|. Therefore the total number of relations we need to convert each fiαito its order form fi′is bounded by
|αi|(mP(2K)kn+m2P2K2n) + (|suppαi| −1)m2(|αi|P)2K2n 6|αi|(mP(2K)kn+ (1+|αi|2)m2P2K2n).
In general, the cost of converting all fiαi’s to their order forms is bounded by
l i=1
∑
|αi|(mP(2K)kn+ (1+|αi|2)m2P2K2n)
=(mP(2K)kn+m2P2K2n)(
l
∑
i|αi|) +m2P2K2n
l i=1
∑
(|αi|3)
6(mP(2K)kn+m2P2K2n)(n2+n)Kn2k+m2P2K2n(n2+n)3K3n2k.
The next step, as described above, is to add all OF(fi)up. We have that|OF(fi)|6|αi||fi|and deg OF(fi)6n for alli=1,2, . . . ,l.Moreover, the size of any partial product∑li=1′ OF(fi),16l′6lis controlled by the following inequalities:
|
l′ i=1
∑
OF(fi)|6
l′ i=1
∑
|αi||fi|6P
l′ i=1
∑
|αi|6P(n2+n)Kn2k,deg(
l′ i=1
∑
αifi)6n.
This is similar to add fi,u’s. By Lemma V.4.1 (a), the cost of the(|l| −1)additions is bounded by
(l−1)m2(P(n2+n)Kn2k)2K2n6(l−1)m2(n2+n)2P2K2n2k+2n.
Now we need to verify the process of those steps above indeed resultw4. This is provided by the factµi=∑lj=1αjµji=
∑lj=1µi j′ and eventually following Lemma V.4.1 we have
l
∏
i=1fiαi=
l
∏
i=1fi′=
l
∏
i=1(aµ1i1′aµ2i2′ . . .aµmim′ ) =
m
∏
j=1a∑
l j=1µi j′
j =aµ11aµ22. . .aµmm=w4.
By our estimation, the cost of the first equality is bounded by(mP(2K)kn+m2P2K2n)(n2+n)Kn2k+m2P2Q2n(n2+ n)3K3n2k and the cost of the third equality is bounded by(l−1)m2(n2+n)2P2K2n2k+2n.Other equalities hold in the free group hence no cost. Therefore
Area(w4) =(n2+n)Kn2k+ (mP(2K)kn+m2P2K2n)(n2+n)Kn2k+m2P2K2n(n2+n)3K3n2k + (l−1)m2(n2+n)2P2K2n2k+2n.
Now we choose a constantC>Klarge enough such that
(n2+n)Kn2k+ (mP(2K)kn+m2P2K2n)(n2+n)Kn2k+m2P2K2n(n2+n)3K3n2k + (l−1)m2(n2+n)2P2K2n2k+2n
6Cn2k
It is clear that suchCexists, for example we can chooseCto be 4m2P2QK. Note thatP,Qonly depends onf1,f2, . . . ,fl, henceR4, and so doesK. ThereforeCis independent ofw.
In conclusion, we start withw=G1 of length at mostn. By converting it four times, we end up with a wordw4, of which area is bounded byCn2k. Thus
w 0 w1 6n2 w2 6(n+n w3 w4.
2)(2K)n 6(n+n2)2K2n
Summing up all the cost fromw1tow4and with the factC>K, we conclude that the area ofwis bounded above by
Area(w)6Cn2k+ (n+n2)2C2n+ (n+n2)(2C)n+n2. This completes the proof.
Chapter VI
Relative Dehn Functions of Finitely Generated Metabelian Groups