Turán’s Theorem Formalization

ro-gut

0.1 Concentrating support on a clique - Improve Operartion

Definition 1 Weight distributions
#

Given a finite simple graph \(G=(V,E)\), a weight distribution is a function

\[ w:V\to \mathbb {R}_{\ge 0} \qquad \text{with}\qquad \sum _{v\in V} w(v)=1. \]

A probability distribution on the vertex Set.

Definition 2 Total edge weight
#

For \(w\in W\), the total edge weight is

\[ f(w)\; :=\; \sum _{e\in E} \text{vp}(w,e). \]

In words: sum the edge values over all edges of the graph.

Definition 3 A better distribution
#

Given a weight function \(w\), define \(w^*\) such that \(\mathrm{supp}(w^*) \subseteq \mathrm{supp}(w)\) and

\[ w.\mathrm{fw} \leq w^*.\mathrm{fw}. \]

That is \(w^*\) is a distribution with non-decreasing total edge weight with the original support of \(w\) preserved.

Definition 4 Improve Operation
#

Given distinct vertices \(v_j \neq v_i\), define \(w' := \texttt{Improve}(W, v_j, v_i)\) by moving all weight from \(v_j\) to \(v_i\):

\[ w'(v_j)=0,\qquad w'(v_i)=w(v_i)+w(v_j),\qquad w'(v)=w(v)\ \text{for } v\notin \{ v_i,v_j\} . \]
Lemma 5 Sum split
#

The sum over edges \(E\) of the function \(\texttt{vp}\) splits as

\[ \sum _{e \in E} \texttt{vp}(w', e) = \sum _{\substack {e \in E \\ v_i \in e}} \texttt{vp}(w', e) + \sum _{\substack {e \in E \\ v_j \in e}} \texttt{vp}(w', e) + \sum _{\substack {e \in E \\ v_i, v_j \notin e}} \texttt{vp}(w', e). \]

That is the total sum decomposes into parts incident to \(v_i\), incident to \(v_j\), and the rest.

Lemma 6 Gain-incidence increases
#

The increment in the sum over edges incident to \(v_i\) is

\[ \sum _{v_i \in e} \texttt{vp}(w', e) - \sum _{v_i \in e} \texttt{vp}(w, e) = \varepsilon \sum _{v_y \in N(v_i)} w(v_y), \]

where \(N(v_i)\) is the neighborhood of \(v_i\).

Lemma 7 Loose-incidence becomes zero
#

The sum over edges incident to \(v_j\) after the transfer satisfies

\[ \sum _{v_j \in e} \texttt{vp}(w', e) = 0. \]
Lemma 8 Unchanged complement
#

For edges \(e\) not incident to \(v_i\) or \(v_j\),

\[ \texttt{vp}(w', e) = \texttt{vp}(w, e). \]

Edges outside \(v_i\) and \(v_j\) neighborhoods remain unchanged under the operation.

Lemma 9 Improve results in non decreaasing distribution
#

Given non adjacent vertices \(v_i\), \(v_j\) and assuming that the neightborsum of weights at \(s_i\) is equal or greater than that of \(s_j\), we have

\[ f(w) \leq f(w'). \]

that is the total edge weight does not decrease after applying Improve.

Lemma 10 Improve strictly reduces support
#

The new distribution \(w'\) has a strictly smaller support:

\[ |\mathrm{supp}(w')| {\lt} |\mathrm{supp}(w)|. \]
Theorem 11 Support of Better is a clique
#

The support of \(w^*\) forms a clique:

\[ \forall v_x, v_y \in \mathrm{supp}(w^*), v_y \neq v_x \implies \{ v_y, v_x\} \in E. \]

In words: every two distinct vertices with positive weight in \(w^*\) are adjacent.

0.2 The Enhance Operation

Definition 12 Enhance Operation
#

Given distinct non-adjacent vertices \(v_j, v_i\), define \(w^+\) by transferring \(\varepsilon {\gt} 0\) weight from \(v_j\) to \(v_i\):

\[ w^+(v_j) = w(v_j) - \varepsilon , \quad w^+(v_i) = W(v_i) + \varepsilon , \quad w^+(v) = W(v) \text{ for } v \neq v_i, v_j, \]

with the condition \(\{ v_j, v_i\} \notin E\). The Enhance transfers weight between non-adjacent vertices to increase edge weight.

Lemma 13 Supported edge partition
#

The edge set \(E\) partitions as

\[ E = E_{v_i} \cup E_{v_j} \cup E_{\mathrm{rest}}, \]

where

\[ E_{v_i} = \{ e \in E : v_i \in e\} , \quad E_{v_j} = \{ e \in E : v_j \in e\} , \quad E_{\mathrm{rest}} = E \setminus (E_{v_i} \cup E_{v_j}). \]

In words: edges are split into the incidence set to \(v_i\), the one to \(v_j\), and the rest.

Lemma 14 Enhance gain sum
#

Under Theorem 12, the change in the sum over edges incident to \(v_i\) satisfies

\[ \sum _{e \in E_{v_i}} \texttt{vp}(w^+, e) - \sum _{e \in E_{v_i}} \texttt{vp}(w, e) = \varepsilon \sum _{v_y \in N(v_i)} w(v_y). \]

That is the gain vertex’s edge contribution increases by \(\varepsilon \) times the sum of its neighbors’ weights.

Lemma 15 Enhance loose sum
#

Under Theorem 12, the sum over edges incident to \(v_j\) satisfies

\[ \sum _{e \in E_{v_j}} \texttt{vp}(w^+, e) = 0. \]

That is the loose vertex’s incident edge contributions become zero after Enhance.

Definition 16 Bijection inside the clique
#

Define a bijection

\[ \phi : \{ e \in E_{v_j} \setminus \{ s(v_j, v_i)\} \} \to \{ e \in E_{v_i} \setminus \{ s(v_j, v_i)\} \} \]

mapping edges incident to \(v_j\) (except \(s(v_j,v_i)\)) to edges incident to \(v_i\) (except \(s(v_j,v_i)\)). In words: this bijection pairs edges incident to \(v_j\) with edges incident to \(v_i\) within the clique.

Lemma 17 Bijection preserves
#

For any edge \(e\) incident to \(v_j\) (excluding \(s(v_j, v_i)\)), the "other" vertex weight satisfies

\[ w(\mathrm{other}(e, v_j)) = w(\mathrm{other}(\phi (e), v_i)). \]

In words: the bijection preserves weights at the other endpoints of edges.

Lemma 18 Loose/gain equality
#

The total weight transfer balances the edge contributions:

\[ \sum _{e \in E_{v_j}} \texttt{vp}(w^+, e) + \sum _{e \in E_{v_i}} \texttt{vp}(w', e) \geq \sum _{e \in E_{v_j}} \texttt{vp}(w, e) + \sum _{e \in E_{v_i}} \texttt{vp}(w, e). \]

In words: the combined edge contributions of loose and gain vertices do not decrease after Enhance.

Lemma 19 Complement unchanged
#

For edges \(e \in E_{\mathrm{rest}}\),

\[ \texttt{vp}(w^+, e) = \texttt{vp}(w, e). \]

In words: edges not incident to \(v_i\) or \(v_j\) remain unaffected by Enhance.

Lemma 20 Edge contribution increase
#

The total edge contribution satisfies:

\[ \sum _{e \in E_{v_i} \cup E_{v_j}} \texttt{vp}(w^+, e) \geq \sum _{e \in E_{v_i} \cup E_{v_j}} \texttt{vp}(w, e). \]

That is, the total contribution from gain and loose vertices does not decrease.

Lemma 21 Support edges unchanged
#

For any vertex \(v \notin \{ v_i, v_j\} \), the edge contributions satisfy

\[ \sum _{e \ni v} \texttt{vp}(w^+, e) = \sum _{e \ni v} \texttt{vp}(w, e). \]

In words: vertices outside gain and loose retain their edge contributions after Enhance.

Theorem 22 Enhance increases edge weight

For a given distribution \(w\in W\) Applying Enhance (\(w^+\)) strictly increases the total edge weight:

\[ f(w^+) {\gt} f(w) \]

That is, the Enhance operation strictly improves the total edge weight contribution.

0.3 Equalizing the weights on the clique - EnhanceD

Definition 23 Maximising the number of uniform vertices
#

For a given distribution \(w\), \(K\) is the maximal number of uniform vertices achievable without decreasing the total edge weight

\[ K := \max \{ N_a(w) \} \]

.

Lemma 24 Best uniform distribution exists
#

There exists \(w_M\) with \(\mathrm{supp}(w_M) \subseteq \mathrm{supp}(w)\), \(w_M.\mathrm{fw} \geq W.\mathrm{fw}\), and with at least \(m\) vertices having weight \(1/m\). In words: a maximiser \(w_M\) achieving the maximal uniform vertex count exists.

Definition 25 UniformBetter
#

Given \(w\in \mathcal{W}\) whose support induces a clique, define

\[ w_M := \texttt{UniformBetter}(w) \]

to be the witness provided by Theorem 24: it preserves the zero set of \(W\), its support is a clique, satisfies \(f(w_M)\ge f(w)\), and achieves the maximal number \(K=\texttt{max\_ uniform\_ support}(w)\) of vertices with weight \(1/k\) (where \(k=|\operatorname {supp}(w)|\)).

Definition 26 Carefully chosen \(\varepsilon \)
#

Define

\[ \mathsf{the\_ \varepsilon } \; :=\; w_{\max } \; -\; \frac{1}{k}. \]

In words: \(\mathsf{the\_ \varepsilon }\) is the difference between the largest vertex weight and the average \(1/k\).

Definition 27 Enhanced Operation
#

Let \(v_{\max }\) and \(v_{\min }\) be vertices attaining the maximal and minimal weights of \(w\), respectively. Set \(\varepsilon := \mathsf{the\_ \varepsilon }\) and define

\[ w^+ \; :=\; \texttt{Enhance}(w, v_{\max }, v_{\min }, \mathsf{the\_ \varepsilon }). \]

In words: \(w^+\) transfers the carefully chosen \(\varepsilon \) from the heaviest to the lightest vertex.

Lemma 28
#

For any vertex \(v\) with \(w(v) = \frac{1}{|\mathrm{supp}(w)|}\),

\[ w^+(v) = w(v). \]

In words: vertices already at uniform weight remain unchanged under Enhanced.

Lemma 29
#

The weight at the argmax vertex \(v_j\) after Enhanced satisfies

\[ w^+(v_j) = \frac{1}{|\mathrm{supp}(w)|}. \]

That is, Enhanced reduces the argmax vertex’s weight to the uniform weight.

Lemma 30
#

The number of vertices with weight \(\frac{1}{|\mathrm{supp}(W)|}\) increases after Enhanced:

\[ |\{ v : w^+(v) = 1/|\mathrm{supp}(w)| \} | {\gt} |\{ v : W(v) = 1/|\mathrm{supp}(w)| \} |. \]
Lemma 31 Uniform weights on the support

For every vertex \(v \in \mathrm{supp}(w_M)\),

\[ w_M(v) = \frac{1}{|\mathrm{supp}(w_M)|}. \]

That is the weights of all support vertices in UniformBetter are uniform.

Lemma 32 Edge values under UniformBetter
#

For any edge \(e = \{ v_a, v_b\} \) with \(v_a, v_b \in \mathrm{supp}(w_M)\),

\[ \texttt{vp}(w_M, e) = \left(\frac{1}{|\mathrm{supp}(w_M)|}\right)^2. \]

In words: every supported edge has value equal to the square of the uniform vertex weight.

Lemma 33 Edge count in a clique
#

If \(|\mathrm{supp}(w_M)| = k\), then

\[ |\{ e \in E : e \subseteq \mathrm{supp}(w_M) \} | = \frac{k(k-1)}{2}. \]
Lemma 34 computation
#

For \(k {\gt} 0\),

\[ \frac{k(k-1)}{2} \cdot \left(\frac{1}{k}\right)^2 = \frac{1}{2}\left(1 - \frac{1}{k}\right). \]

That is the total edge weight for a clique with uniform weights simplifies to \(\frac{1}{2}(1 - \frac{1}{k})\).

Lemma 35 Monotonicity of the bound
#

The function

\[ f(k) := \frac{1}{2}\left(1 - \frac{1}{k}\right) \]

is nondecreasing for \(k \geq 1\).

Theorem 36 Final bound inside a clique

If \(w\) is supported on a clique of size \(k \leq p-1\), then

\[ f(w) \leq \frac{1}{2}\left(1 - \frac{1}{p-1}\right). \]

In words: the total edge weight is bounded by the Turán bound for cliques of size less than \(p\).

Definition 37 Uniform weights over all vertices
#

Define

\[ \texttt{UnivFun}(G)(v) := \frac{1}{|V|} \quad \forall v \in V. \]

That is, the uniform vertex weight distribution assigns equal weight \(1/|V|\) to every vertex.

Lemma 38 Total weight under \(\texttt{UnivFun}\)
#

The total edge weight satisfies

\[ (\texttt{UnivFun}(G)).\mathrm{fw} = |E| \cdot \left(\frac{1}{|V|}\right)^2. \]

THat is, the total edge weight under uniform vertex weights equals the number of edges times the square of the uniform weight.

Theorem 39 Turán’s Theorem
#

Let \(p \geq 2\) and let \(G\) be a \(p\)-clique-free graph. Then

\[ |E| \leq \frac{1}{2}\left(1 - \frac{1}{p-1}\right) |V|^2. \]