Additive Combinatorics and its Applications in Theoretical CS

Additive Combinatorics
and its Applications in
Theoretical CS
By Shachar Lovett
Chapter 2: Set addition, Sections 2.1-2.3, pp. 3-8.
Presented by Tomer Bincovich
Set addition
β€’ 𝐺 an abelian group (think about 𝐺 = 𝔽𝑛 ).
β€’ 𝐴 βŠ† 𝐺.
β€’ The sumset of 𝐴 is
2𝐴 = 𝐴 + 𝐴 = π‘Ž + π‘Žβ€² π‘Ž, π‘Žβ€² ∈ 𝐴
β€’ Always 2𝐴 β‰₯ 𝐴 .
β€’ When does equality hold?
When does 2𝐴 = |𝐴|?
β€’ Equality holds if 𝐴 is empty, a subgroup of 𝐺 or a coset of a subgroup.
β€’ For the other direction:
β€’ If 𝐴 β‰  βˆ…, we can assume w.l.o.g. 0 ∈ 𝐴 by shifting.
β€’ Then 𝐴 βŠ† 2𝐴, and since 2𝐴 = 𝐴 we have that 2𝐴 = 𝐴.
β€’ 𝐴 is a nonempty finite subset that is closed under addition, hence is a
subgroup.
Planned topics
β€’ Ruzsa calculus
β€’ The span of sets of small doubling
β€’ The growth of sets of small doubling
Ruzsa calculus
β€’ A set of basic inequalities between sizes of sets and their sumsets.
β€’ For 𝐴, 𝐡 βŠ† 𝐺, define:
β€’ Sumset: 𝐴 + 𝐡 = π‘Ž + 𝑏 π‘Ž ∈ 𝐴, 𝑏 ∈ 𝐡
β€’ Difference set: 𝐴 βˆ’ 𝐡 = π‘Ž βˆ’ 𝑏 π‘Ž ∈ 𝐴, 𝑏 ∈ 𝐡
Claim 2.1 (Ruzsa triangle inequality):
𝐴 π΅βˆ’πΆ ≀ π΄βˆ’π΅ π΄βˆ’πΆ
Proof of Claim 2.1
β€’ Claim 2.1 (Ruzsa triangle inequality):
𝐴 π΅βˆ’πΆ ≀ π΄βˆ’π΅ π΄βˆ’πΆ
β€’ Define a map 𝑓: 𝐴 × π΅ βˆ’ 𝐢 β†’ 𝐴 βˆ’ 𝐡 × π΄ βˆ’ 𝐢 :
for any π‘₯ ∈ 𝐡 βˆ’ 𝐢 fix 𝑏 ∈ 𝐡, 𝑐 ∈ 𝐢 s.t. π‘₯ = 𝑏 βˆ’ 𝑐, and define
𝑓 π‘Ž, π‘₯ = π‘Ž βˆ’ 𝑏, π‘Ž βˆ’ 𝑐 .
β€’ 𝑓 is injective, hence 𝐴 𝐡 βˆ’ 𝐢 ≀ 𝐴 βˆ’ 𝐡 𝐴 βˆ’ 𝐢 .
The Ruzsa distance
π΄βˆ’π΅
𝑑 𝐴, 𝐡 = log 1 2 1
𝐴
𝐡
β€’ Not formally a distance function. Why?
2
β€’ Claim 2.3: The Ruzsa distance is symmetric and obeys the triangle
inequality.
Proof of Claim 2.3
π΄βˆ’π΅
𝑑 𝐴, 𝐡 = log 1 2 1
𝐴
𝐡
β€’ Symmetry: since 𝐡 βˆ’ 𝐴 = 𝐴 βˆ’ 𝐡 .
β€’ Moreover, 𝑑 𝐴, 𝐢 ≀ 𝑑 𝐴, 𝐡 + 𝑑 𝐡, 𝐢 is equivalent to
π΄βˆ’πΆ
π΄βˆ’π΅
π΅βˆ’πΆ
log 1 2 1 2 ≀ log 1 2 1 2 + log 1 2 1
𝐴
𝐢
𝐴
𝐡
𝐡
𝐢
π΄βˆ’πΆ
𝐴1 2𝐢1
2
π΄βˆ’π΅
≀
𝐴1 2𝐡1
2
π΅βˆ’πΆ
𝐡1 2𝐢1
𝐡 π΄βˆ’πΆ ≀ π΅βˆ’π΄ π΅βˆ’πΆ
Which follows from Claim 2.1.
2
2
2
Corollary 2.4
If 𝐴 βˆ’ 𝐡 ≀ 𝐾 𝐴
1 2
𝐡
1 2
then 𝐴 βˆ’ 𝐴 ≀ 𝐾 2 𝐴 .
β€’ For 𝐡 = βˆ’π΄ we get that if 𝐴 + 𝐴 ≀ 𝐾 𝐴 then 𝐴 βˆ’ 𝐴 ≀ 𝐾 2 𝐴 .
Proof of Corollary 2.4
If 𝐴 βˆ’ 𝐡 ≀ 𝐾 𝐴
1 2
π΄βˆ’π΄
= 𝑒𝑑
𝐴
𝐡
𝐴,𝐴
1 2
≀𝑒
then 𝐴 βˆ’ 𝐴 ≀ 𝐾 2 𝐴 .
𝑑 𝐴,𝐡 +𝑑 𝐡,𝐴
=
π΄βˆ’π΅
𝐴1 2𝐡1
π΄βˆ’π΅
𝑑 𝐴, 𝐡 = log 1 2 1
𝐴
𝐡
2
2
2
≀ 𝐾2
The doubling constant
β€’ The doubling constant of 𝐴 is 𝐾 =
𝐴+𝐴
𝐴
.
β€’ We saw that 𝐾 = 1 corresponds to subgroups.
β€’ Can we use 𝐾 to say something about the size of the smallest
subgroup containing 𝐴? (or, in the case of 𝐺 = 𝔽𝑛 , the size of the
span of 𝐴)
Lemma 2.5 (Laba)
If 𝐴 βˆ’ 𝐴 <
3
2
𝐴 then 𝐴 βˆ’ 𝐴 is a subgroup.
β€’ The constant 3/2 is tight: take 𝐴 = 0,1 βŠ† β„€.
𝐴 βˆ’ 𝐴 = βˆ’1,0,1 is not a subgroup.
Proof of Lemma 2.5
β€’ We will show that for any π‘₯ ∈ 𝐴 βˆ’ 𝐴, 𝐴 ∩ 𝐴 + π‘₯ > 𝐴 /2.
β€’ This implies that for any π‘₯, 𝑦 ∈ 𝐴 βˆ’ 𝐴, 𝐴 + π‘₯ ∩ 𝐴 + 𝑦 β‰  βˆ…:
β€’ Denoting 𝐴π‘₯ = 𝐴 ∩ 𝐴 + π‘₯ , by the inclusion exclusion principle
𝐴+π‘₯ ∩ 𝐴+𝑦
β‰₯ 𝐴π‘₯ ∩ 𝐴𝑦 = 𝐴π‘₯ + 𝐴 𝑦 βˆ’ 𝐴π‘₯ βˆͺ A 𝑦
> 𝐴 /2 + 𝐴 /2 βˆ’ 𝐴 = 0
β€’ There are π‘Ž1 , π‘Ž2 ∈ 𝐴 s.t. π‘Ž1 + π‘₯ = π‘Ž2 + 𝑦, thus π‘₯ βˆ’ 𝑦 = π‘Ž2 βˆ’ π‘Ž1
∈ 𝐴 βˆ’ 𝐴.
β€’ 𝐴 βˆ’ 𝐴 is closed under taking difference, hence must be a subgroup.
Proof of Lemma 2.5 cont.
β€’ We will show that for any π‘₯ ∈ 𝐴 βˆ’ 𝐴, 𝐴 ∩ 𝐴 + π‘₯ > 𝐴 /2.
β€’ Let π‘₯ = π‘Ž βˆ’ π‘Žβ€² ∈ 𝐴 βˆ’ 𝐴. Then 𝐴 ∩ 𝐴 + π‘₯ = 𝐴 βˆ’ π‘Ž ∩ 𝐴 βˆ’ π‘Žβ€² .
β€’ Similarly to before:
𝐴 βˆ’ π‘Ž ∩ 𝐴 βˆ’ π‘Žβ€²
= 𝐴 βˆ’ π‘Ž + 𝐴 βˆ’ π‘Žβ€² βˆ’ 𝐴 βˆ’ π‘Ž βˆͺ 𝐴 βˆ’ π‘Žβ€²
β‰₯ 𝐴 + 𝐴 βˆ’ 𝐴 βˆ’ 𝐴 > |𝐴|/2
As needed.
Lemma 2.6 (Freiman)
Let 𝐴 βŠ† ℝ𝑛 with 𝐴 + 𝐴 ≀ 𝐾 𝐴 . Then 𝐴 lies in an affine subspace of
dimension at most 2𝐾 βˆ’ 1.
Proof of Lemma 2.6
β€’ We will prove that if 𝐴 has affine dimension 𝑑 then
𝑑+1
𝐴+𝐴 β‰₯ 𝑑+1 𝐴 βˆ’
2
β€’ Thus 𝑑 + 1 𝐴 βˆ’
𝑑+1
2
≀ 𝐴 + 𝐴 ≀ 𝐾 𝐴 , and since 𝐴 β‰₯ 𝑑,
𝑑+1
𝑑+1
𝑑+1βˆ’πΎ 𝑑 ≀ 𝑑+1βˆ’πΎ 𝐴 ≀
=
𝑑
2
2
β€’ Hence 𝐾 β‰₯
𝑑+1
,
2
or 𝑑 ≀ 2𝐾 βˆ’ 1.
Proof of Lemma 2.6 cont.
β€’ We prove by induction on 𝐴 that if 𝐴 has affine dimension 𝑑 then
𝑑+1
𝐴+𝐴 β‰₯ 𝑑+1 𝐴 βˆ’
2
β€’ If 𝐴 = 1 then 𝐴 + 𝐴 = 1, 𝑑 = 0 and the claim follows.
π‘Ž+π‘Žβ€²
2
β€’ Assume 𝐴 > 1. Let 𝑀 𝐴 =
π‘Ž, π‘Žβ€² ∈ 𝐴 .
Clearly 𝑀 𝐴 = 𝐴 + 𝐴 .
β€’ Let 𝐢(𝐴) be the convex hull of 𝐴, which is a 𝑑-dimensional polytope
by assumption. Its vertices are the points in 𝐴 that cannot be
obtained as a convex combination of the other points.
Inductive proof, case (1)
β€’ Fix a vertex π‘Žβˆ— ∈ 𝐴 of 𝐢 𝐴 , and let 𝐴′ = 𝐴 βˆ– π‘Žβˆ— . We have two cases:
π‘Ž+π‘Žβˆ—
2
β€²
1) The affine dimension of 𝐴 is 𝑑 βˆ’ 1. Then the mid-points
for all
π‘Ž ∈ 𝐴 are outside 𝐢(𝐴′ ).
𝑑
𝑑
β€²
β€²
𝑀 𝐴 β‰₯ 𝐴 + 𝑀 𝐴 β‰₯ 𝐴 +𝑑 𝐴 βˆ’
= 𝑑+1 𝐴 βˆ’π‘‘βˆ’
2
2
𝑑+1
= 𝑑+1 𝐴 βˆ’
2
?
𝑑+1
𝐴+𝐴 β‰₯ 𝑑+1 𝐴 βˆ’
2
Inductive proof, case (2)
β€’ Fix a vertex π‘Žβˆ— ∈ 𝐴 of 𝐢 𝐴 , and let 𝐴′ = 𝐴 βˆ– π‘Žβˆ— . We have two cases:
2) The affine dimension of 𝐴′ is 𝑑. Let π‘Ž1 , … , π‘Žπ‘‘ be the neighboring vertices
to π‘Žβˆ— in 𝐢 𝐴 (two vertices are neighbors if the segment connecting
them is a one-dim. face of 𝐢 𝐴 ). Note that 𝑑 β‰₯ 𝑑 since 𝐢 𝐴 is 𝑑-dim.
π‘Žβˆ— +π‘Žπ‘–
βˆ—
The points π‘Ž and
for 1 ≀ 𝑖 ≀ 𝑑 are mid-points outside 𝐢 𝐴′ .
2
𝑑+1
β€²
β€²
𝑀 𝐴 β‰₯ 𝑑+1 + 𝑀 𝐴 β‰₯𝑑+1+ 𝑑+1 𝐴 βˆ’
2
𝑑+1
= 𝑑+1 𝐴 βˆ’
?
𝑑+1
2
𝐴+𝐴 β‰₯ 𝑑+1 𝐴 βˆ’
2
Lemma 2.6 – tightness
Let 𝐴 βŠ† ℝ𝑛 with 𝐴 + 𝐴 ≀ 𝐾 𝐴 . Then 𝐴 lies in an affine subspace of
dimension at most 2𝐾 βˆ’ 1.
β€’ The dimension is tight up to lower order terms. Take 𝐴 = 𝑣1 , … , 𝑣2𝐾
a set of linearly independent vectors, then 𝐴 + 𝐴 = 2𝐾
+ 2𝐾
2
𝐴+𝐴
1
= 𝐾(2𝐾 + 1) and
=𝐾+ .
𝐴
2
Theorem 2.7 (Ruzsa)
Let 𝐺 be an Abelian group of torsion π‘Ÿ, 𝐴 βŠ† 𝐺 with 𝐴 + 𝐴 ≀ 𝐾 𝐴 .
Then there exists a subgroup 𝐻 < 𝐺, 𝐻 ≀
4
2
𝐾
𝐾 π‘Ÿ
𝐴 s.t. 𝐴 βŠ† 𝐻.
β€’ A group has torsion π‘Ÿ β‰₯ 1 if π‘Ÿ βˆ™ 𝑔 = 0 for all 𝑔 ∈ 𝐺.
Example
β€’ 𝐺 = 𝔽𝑛𝑝 has torsion π‘Ÿ = 𝑝. Let π‘ˆ, 𝑉 βŠ† 𝔽𝑛𝑝 two subspaces with
π‘ˆ ∩ 𝑉 = 0 , and set 𝐴 = π‘ˆ + {𝑣1 , … , 𝑣2𝐾 } where 𝑣1 , … , 𝑣2𝐾 ∈ 𝑉 are
linearly independent.
β€’ 𝐴 = π‘ˆ βˆ™ 2𝐾. Since 𝐴 + 𝐴 = π‘ˆ + 𝑣𝑖 + 𝑣𝑗 1 ≀ 𝑖 ≀ 𝑗 ≀ 2𝐾 ,
𝐴+𝐴
1
𝐴 + 𝐴 = π‘ˆ βˆ™ 𝐾 2𝐾 + 1 and
= 𝐾 + β‰ˆ 𝐾.
𝐴
2
β€’ However, the size of the minimal subspace containing 𝐴 is
2𝐾
𝑝
𝑝2𝐾 π‘ˆ =
𝐴
2𝐾
β€’ Thus an exponential dependency on 𝐾 is unavoidable.
Conjecture 2.8 (Ruzsa)
There exists an absolute constant 𝐢 β‰₯ 2 s.t. for any Abelian group 𝐺 of
torsion π‘Ÿ, and any 𝐴 βŠ† 𝐺 with 𝐴 + 𝐴 ≀ 𝐾|𝐴|, the subgroup generated
by 𝐴 has order ≀ π‘Ÿ 𝐢𝐾 𝐴 .
β€’ Was established with 𝐢 = 2 for π‘Ÿ = 2 and then extended for prime π‘Ÿ:
β€’ Theorem 2.9:
Let 𝑝 be a prime, 𝐴 βŠ† 𝔽𝑛𝑝 with 𝐴 + 𝐴 ≀ 𝐾 𝐴 . Then there exists a
subspace 𝐻 βŠ† 𝔽𝑛𝑝 s.t. 𝐴 βŠ† 𝐻 and 𝐻 ≀
𝑝2𝐾
2πΎβˆ’1
𝐴.
Theorem 2.10
If 2𝐴 ≀ 𝐾 𝐴 then 2𝐴 βˆ’ 2𝐴 ≀ 𝐾 4 𝐴 .
β€’ Used to prove Theorem 2.7.
β€’ We will prove a more general result later today.
Proof of Theorem 2.7 - Sketch
β€’ Let 𝐴 βŠ† 𝐺 with 2𝐴 ≀ 𝐾 𝐴 . We can assume 0 ∈ 𝐴 by possibly
replacing 𝐴 with 𝐴 βˆ’ π‘Ž for some π‘Ž ∈ 𝐴.
β€’ Let 𝐡 = 𝑏1 , … , π‘π‘š βŠ† 2𝐴 βˆ’ 𝐴 be a maximal collection of elements
s.t. 𝑏𝑖 βˆ’ 𝐴 are all disjoint (β€œRuzsa covering”).
β€’ 𝐴 has small doubling, thus Theorem 2.10 implies that 𝐡 is bounded.
β€’ We will show that ℓ𝐴 βˆ’ 𝐴 βŠ† β„“ βˆ’ 1 𝐡 + 𝐴 βˆ’ 𝐴 for all β„“ β‰₯ 1.
β€’ Together the theorem follows.
Proof of Theorem 2.7 cont.
β€’ Since 𝑏𝑖 βˆ’ 𝐴 βŠ† 2𝐴 βˆ’ 2𝐴,
2𝐴 βˆ’ 2𝐴
π‘šβ‰€
≀ 𝐾4
𝐴
β€’ Therefore 𝐡 ≀ 𝐾 4 .
2𝐴 ≀ 𝐾 𝐴 β‡’ 2𝐴 βˆ’ 2𝐴 ≀ 𝐾 4 𝐴
ℓ𝐴 βˆ’ 𝐴 βŠ† β„“ βˆ’ 1 𝐡 + 𝐴 βˆ’ 𝐴
β€’ β„“ = 1 is trivial.
β€’ For β„“ = 2, we need to show 2𝐴 βˆ’ 𝐴 βŠ† 𝐡 + 𝐴 βˆ’ 𝐴.
β€’ Take π‘₯ ∈ 2𝐴 βˆ’ 𝐴, by construction there exists 𝑏 ∈ 𝐡 s.t.
π‘₯ βˆ’ 𝐴 ∩ 𝑏 βˆ’ 𝐴 β‰  βˆ…, i.e. π‘₯ βˆ’ π‘Ž = 𝑏 βˆ’ π‘Žβ€² for some π‘Ž, π‘Žβ€² ∈ 𝐴.
β€’ Hence π‘₯ = 𝑏 + π‘Ž βˆ’ π‘Žβ€² ∈ 𝐡 + 𝐴 βˆ’ 𝐴.
β€’ By induction on β„“,
ℓ𝐴 βˆ’ 𝐴 = 𝐴 + β„“ βˆ’ 1 𝐴 βˆ’ 𝐴 βŠ† 𝐴 + β„“ βˆ’ 2 𝐡 + 𝐴 βˆ’ 𝐴
βŠ† β„“βˆ’2 𝐡+𝐡+π΄βˆ’π΄ = β„“βˆ’1 𝐡+π΄βˆ’π΄
Concluding the proof
ℓ𝐴 βˆ’ 𝐴 βŠ† β„“ βˆ’ 1 𝐡 + 𝐴 βˆ’ 𝐴
β€’ Let 𝐴 be the subgroup spanned by 𝐴, and similarly for 𝐡.
𝐴 =
ℓ𝐴 βŠ†
β„“β‰₯1
ℓ𝐴 βˆ’ 𝐴 βŠ†
β„“β‰₯1
β„“βˆ’1 𝐡+π΄βˆ’π΄ = 𝐡 +π΄βˆ’π΄
β„“β‰₯1
β€’ Hence, using Corollary 2.4 (which implies 𝐴 βˆ’ 𝐴 ≀ 𝐾 2 𝐴 ),
𝐴 ≀ 𝐡 βˆ™ 𝐴 βˆ’ 𝐴 ≀ 𝐡 βˆ™ 𝐾2 𝐴
β€’ We conclude by bounding 𝐡 . As 𝐺 has torsion π‘Ÿ and we saw that
4
4
𝐾
𝐡 ≀ 𝐾 we have 𝐡 ≀ π‘Ÿ .
β€’ Finally, 𝐴 ≀
4
2
𝐾
𝐾 π‘Ÿ
𝐴.
Iterated sumsets
ℓ𝐴 = π‘Ž1 + β‹― + π‘Žβ„“ π‘Ž1 , … , π‘Žβ„“ ∈ 𝐴
ℓ𝐴 βˆ’ π‘šπ΄ = π‘Ž1 + β‹― + π‘Žβ„“ βˆ’ π‘Žβ„“+1 βˆ’ β‹― βˆ’ π‘Žβ„“+π‘š π‘Ž1 , … , π‘Žβ„“+π‘š ∈ 𝐴
β€’ Theorem 2.11 (Plünneke-Ruzsa): If 𝐴 = 𝐡 and 𝐴 + 𝐡 ≀ 𝐾 𝐴
then ℓ𝐴 βˆ’ π‘šπ΄ ≀ 𝐾 β„“+π‘š 𝐴 .
β€’ For 𝐡 = 𝐴 or 𝐡 = βˆ’π΄ we obtain the following:
β€’ Corollary 2.12: If 𝐴 + 𝐴 ≀ 𝐾 𝐴 or 𝐴 βˆ’ 𝐴 ≀ 𝐾 𝐴
then ℓ𝐴 βˆ’ π‘šπ΄ ≀ 𝐾 β„“+π‘š 𝐴 .
Lemma 2.13
Let 𝐴, 𝐡 βŠ† 𝐺 with 𝐴 = 𝐡 and 𝐴 + 𝐡 ≀ 𝐾 𝐴 . Let 𝐡0 βŠ† 𝐡 a
𝐴+𝐡0
nonempty set minimizing the ratio 𝐾0 =
. Then for any 𝐢 βŠ† 𝐺,
𝐡0
𝐴 + 𝐡0 + 𝐢 ≀ 𝐾0 𝐡0 + 𝐢
β€’ Notice that 𝐾0 ≀ 𝐾.
Proof of Theorem 2.11
β€’
β€’
β€’
β€’
Let 𝐡0 βŠ† 𝐡 as in Lemma 2.13. βˆ€πΆ 𝐴 + 𝐡0 + 𝐢 ≀ 𝐾0 𝐡0 + 𝐢
We will prove by induction on β„“ that 𝐡0 + ℓ𝐴 ≀ 𝐾0β„“ 𝐡0 ≀ 𝐾 β„“ 𝐡0 .
β„“ = 0 is easy: 𝐡0 + 0𝐴 = 1 βˆ™ 𝐡0 .
For β„“ > 0:
𝐡0 + ℓ𝐴 = 𝐴 + 𝐡0 + (β„“ βˆ’ 1)𝐴 ≀ 𝐾0 𝐡0 + β„“ βˆ’ 1 𝐴
≀ 𝐾0 βˆ™ 𝐾0β„“βˆ’1 𝐡0 = 𝐾0β„“ 𝐡0
β€’ By the Ruzsa triangle inequality (Claim 2.1), applied to βˆ’π΅0 , ℓ𝐴, π‘šπ΄:
βˆ’π΅0 ℓ𝐴 βˆ’ π‘šπ΄ ≀ 𝐡0 + ℓ𝐴 𝐡0 + π‘šπ΄ ≀ 𝐾 β„“+π‘š 𝐡0 2
𝐴 π΅βˆ’πΆ ≀ π΄βˆ’π΅ π΄βˆ’πΆ
β€’ Hence
ℓ𝐴 βˆ’ π‘šπ΄ ≀ 𝐾 β„“+π‘š 𝐡0 ≀ 𝐾 β„“+π‘š 𝐴
Lemma 2.13 (Reminder)
Let 𝐴, 𝐡 βŠ† 𝐺 with 𝐴 = 𝐡 and 𝐴 + 𝐡 ≀ 𝐾 𝐴 . Let 𝐡0 βŠ† 𝐡 a
𝐴+𝐡0
nonempty set minimizing the ratio 𝐾0 =
. Then for any 𝐢 βŠ† 𝐺,
𝐡0
𝐴 + 𝐡0 + 𝐢 ≀ 𝐾0 𝐡0 + 𝐢
Proof of Lemma 2.13
β€’ Observe that by minimality of 𝐾0 , for any 𝐡′ βŠ† 𝐡 we have that
𝐴 + 𝐡′ β‰₯ 𝐾0 𝐡′ .
β€’ The proof is by induction on 𝐢 .
β€’ If 𝐢 = π‘₯ ,
𝐴 + 𝐡0 + π‘₯ = 𝐴 + 𝐡0 = 𝐾0 𝐡0 = 𝐾0 𝐡0 + π‘₯
β€’ If 𝐢 > 1, write 𝐢 = 𝐢 β€² βˆͺ π‘₯ .
β€’ Define 𝐡′ βŠ† 𝐡0 as 𝐡′ = 𝑏 ∈ 𝐡0 𝐴 + 𝑏 + π‘₯ βŠ† 𝐴 + 𝐡0 + 𝐢 β€² .
β€’ We can write
𝐴 + 𝐡0 + 𝐢 = 𝐴 + 𝐡0 + 𝐢 β€² βˆͺ 𝐴 + 𝐡0 + π‘₯ βˆ– 𝐴 + 𝐡′ + π‘₯
Proof of Lemma 2.13 cont.
𝐴 + 𝐡0 + 𝐢 = 𝐴 + 𝐡0 + 𝐢 β€² βˆͺ
𝐴 + 𝐡0 + π‘₯ βˆ– 𝐴 + 𝐡′ + π‘₯
𝐴 + 𝐡0 + 𝐢 ≀ 𝐴 + 𝐡0 + 𝐢 β€² + 𝐴 + 𝐡0 + π‘₯ βˆ– 𝐴 + 𝐡′ + π‘₯
= 𝐴 + 𝐡0 + 𝐢 β€² + 𝐴 + 𝐡0 + π‘₯ βˆ’ 𝐴 + 𝐡′ + π‘₯
≀ 𝐾0 𝐡0 + 𝐢 β€² + 𝐴 + 𝐡0 βˆ’ 𝐴 + 𝐡′ βˆ€π΅β€² βŠ† 𝐡
β€² β‰₯ 𝐾 𝐡′
β€²
β€²
𝐴
+
𝐡
0
≀ 𝐾0 𝐡0 + 𝐢 + 𝐾0 𝐡0 βˆ’ 𝐾0 𝐡
= 𝐾0 𝐡0 + 𝐢 β€² + 𝐡0 βˆ’ 𝐡′
β€’ It remains to show that:
𝐡0 + 𝐢 β€² + 𝐡0 βˆ’ 𝐡′ ≀ 𝐡0 + 𝐢
Concluding the proof
𝐡′ = 𝑏 ∈ 𝐡0 𝐴 + 𝑏 + π‘₯ βŠ† 𝐴 + 𝐡0 + 𝐢 β€²
β€’ It remains to show that:
𝐡0 + 𝐢 β€² + 𝐡0 βˆ’ 𝐡′ ≀ 𝐡0 + 𝐢
β€’ Define 𝐡′′ = 𝑏 ∈ 𝐡0 𝑏 + π‘₯ ∈ 𝐡0 + 𝐢 β€² . Note that 𝐡′′ βŠ† 𝐡′ .
β€’ We decompose 𝐡0 + 𝐢 as a disjoint union:
𝐡0 + 𝐢 = 𝐡0 + 𝐢 β€² βˆͺ 𝐡0 + π‘₯ βˆ– 𝐡′′ + π‘₯
β€’ Thus
𝐡0 + 𝐢 = 𝐡0 + 𝐢 β€² + 𝐡0 βˆ’ 𝐡′′ β‰₯ 𝐡0 + 𝐢 β€² + 𝐡0 βˆ’ 𝐡′
Which is what we needed to show.