--- title: Chung–Graham Representation url: https://www.emergentmind.com/topics/chung-graham-representation type: topic --- # Chung–Graham Representation Searching arXiv for papers explicitly about “Chung-Graham representation” and closely related usages. First, I’ll look for papers with “Chung-Graham” in the title or abstract. The term **Chung–Graham representation** is used in several technically distinct senses associated with work of Fan Chung and Ronald Graham. In extremal combinatorics, it denotes an organized hierarchy of quasirandomness properties, recording implications, equivalences, and separations among formal pseudorandomness tests. In algebraic and enumerative graph theory, it refers to classical expansions of chromatic polynomials in monomial and falling-factorial bases, together with gain-graph generalizations. In enumerative combinatorics on permutations, it denotes a polynomial representation of bounded-drop descent distributions through the families \(P_k(x)\), \(Q_k(x)\), and \(R_{n,k}(x)\). In additive numeration, it denotes the unique decomposition of integers into even-indexed Fibonacci numbers with digits in \(\{0,1,2\}\) subject to a spacing rule for the digit \(2\). These usages share a common structural feature: each provides a canonical representation that converts a class of objects into a constrained algebraic, combinatorial, or logical form [1208.5978], [2504.00314], [1306.5428]. ## 1. Quasirandomness hierarchies in graphs and hypergraphs In the quasirandomness literature, the Chung–Graham representation organizes a family of quasirandom properties into a logically determined hierarchy: each property is a formal test for pseudorandom behavior, and the representation consists of the implications, equivalences, and separations among them [1208.5978]. For graphs, Chung–Graham–Wilson showed that many natural dense quasirandomness properties are equivalent. If \(G\) is a graph on \(n\) vertices with edge density \(p=e(G)/\binom{n}{2}\), the equivalent properties include discrepancy on all induced subgraphs, \[ |E(G[U])| = p \binom{|U|}{2} \pm o(n^2), \] cut or expansion estimates \[ E_G(S,T) = p |S| |T| \pm o(n^2), \] small subgraph counts such as \[ \operatorname{inj}(F,G) = p^{e(F)} n^{v(F)} \pm o(n^{v(F)}), \] and spectral control \[ \lambda_2(A)=o(n), \] where \(A\) is the adjacency matrix [1208.5978]. For \(k\)-uniform hypergraphs with \(k \ge 3\), the situation fragments. The paper "The Poset of Hypergraph Quasirandomness" determines the full poset of implications among the principal quasirandom properties—expansion, clique discrepancy, and deviation—thereby resolving Chung’s problem on the quasirandom hierarchy for hypergraphs [1208.5978]. A \(k\)-uniform hypergraph \(H\) has edge density \(p=e(H)/\binom{n}{k}\), and the relevant properties include: - \(\mathrm{Expand}_{\pi}\), indexed by a proper partition \(\pi=k_1+\cdots+k_t\) of \(k\), requiring \[ e(S_1,\dots,S_t)=p \prod_{i=1}^t |S_i| + o(n^k). \] - \(\mathrm{CliqueDisc}_{\ell}\), requiring \[ |E(H_n)\cap \mathcal{K}_k(G)| = p |\mathcal{K}_k(G)| + o(n^k) \] for every \(\ell\)-uniform \(G\). - \(\mathrm{Deviation}_{\ell}\), defined by signed octahedral parity sums of order \(o(n^{k+\ell})\) [1208.5978]. The representation theorem establishes the exact logical structure. For proper partitions \(\pi,\pi'\) of \(k\), \[ \mathrm{Expand}_{\pi}\Rightarrow \mathrm{Expand}_{\pi'} \quad\text{iff}\quad \pi' \text{ is a refinement of } \pi. \] Also, \[ \mathrm{CliqueDisc}_{\ell}\Rightarrow \mathrm{Expand}_{\pi} \quad\text{iff}\quad \max_i k_i \le \ell, \] while every \(\mathrm{Expand}_{\pi}\) implies \(\mathrm{CliqueDisc}_1\) but no \(\mathrm{Expand}_{\pi}\) implies \(\mathrm{CliqueDisc}_{\ell}\) for \(\ell \ge 2\). The deviation properties sit above all expansions: \[ \mathrm{Deviation}_{\ell}\Rightarrow \mathrm{Expand}_{\pi} \] for all \(\pi\), and \(\mathrm{Deviation}_{k}\) is equivalent to \(\mathrm{CliqueDisc}_{k-1}\) [1208.5978]. A central significance of this representation is the contrast with the graph case. For graphs, one certificate effectively certifies all the others; for hypergraphs, the representation is a genuine poset rather than a single equivalence class. This suggests that hypergraph quasirandomness is intrinsically multi-level rather than unified. The same paper notes consequences for testing and embedding: to certify \(\mathrm{Expand}_{\pi}\), it suffices to count \(\pi\)-linear subhypergraphs asymptotically, while clique-based tests certify expansions only up to the largest block size \(\max \pi\) [1208.5978]. ## 2. Oriented-graph and tournament analogues A closely related use of the Chung–Graham representation occurs for tournaments and oriented graphs. The paper "Quasi-random oriented graphs" extends Chung–Graham’s tournament characterization to orientations of an arbitrary underlying graph \(G\) [1101.5421]. Here the representation is again a system of equivalent quasirandomness conditions, now expressed relative to a fixed undirected host graph. Let \(D\) be an oriented graph with underlying graph \(G\), and let \(M\) be its skew-adjacency matrix, with entries \(M_{xy}=1\) if \(xy\in E(D)\), \(M_{xy}=-1\) if \(yx\in E(D)\), and \(M_{xy}=0\) otherwise. The main theorem proves equivalence among conditions including: - correct counts of certain oriented \(4\)-cycle types, such as \[ | \mathrm{hom}_{IV}(D) - (1/8)\mathrm{hom}(C_4,G) | \le \alpha n^4; \] - correct counts of all small oriented patterns, \[ |\mathrm{hom}(H,D)-2^{-e(H)}\mathrm{hom}(\bar H,G)| \le \beta n^k; \] - uniform cut-direction balance, \[ |e_D(A\to B)-e_D(B\to A)| \le \gamma n; \] - fourth-moment and spectral conditions on \(M\), including \[ \sum_{i=1}^n |\lambda_i|^4 \le \delta n^4 \quad\text{and}\quad |\lambda_1|\le \zeta n \] [1101.5421]. One notable feature is that exactly two of the four orientations of a \(4\)-cycle can serve as quasirandomness certificates. Type IV cycles satisfy the lower bound \[ \mathrm{hom}_{IV}(D)\ge (1/8)\mathrm{hom}(C_4,G), \] and Type II cycles satisfy the upper bound \[ \mathrm{hom}_{II}(D)\le (1/2)\mathrm{hom}(C_4,G). \] If either count is close to its random-orientation benchmark, then all the other quasirandom properties follow [1101.5421]. This oriented-graph representation is not the same object as the hypergraph poset. It is instead an equivalence theorem in the style of Chung–Graham–Wilson, but adapted to orientations and normalized by the underlying graph \(G\). A plausible implication is that “Chung–Graham representation” in quasirandomness should be understood not as a single theorem but as a program: encode pseudorandomness by a family of test properties and determine their exact logical relations. ## 3. Chromatic-polynomial expansions and gain graphs In chromatic graph theory, the Chung–Graham representation refers to basis expansions of the chromatic polynomial, especially the monomial basis \(q^k\) and the falling-factorial basis \((q)_k\). For an ordinary graph \(G\), the classical identities are \[ P_G(q)=\sum_{S\subseteq E(G)}(-1)^{|S|}q^{c(S)} \] and \[ P_G(q)=\sum_{\pi\in \Pi^*(G)} (q)_{|\pi|}, \qquad (q)_m=q(q-1)\cdots(q-m+1), \] where \(\Pi^*(G)\) denotes the stable partitions of \(V(G)\) [1001.4216]. The second formula is the Chung–Graham-type falling-factorial representation: coefficients count stable partitions, and the basis terms come from complete graphs on the contracted blocks. The paper "An elementary chromatic reduction for gain graphs and special hyperplane arrangements" generalizes these expansions to gain graphs through the neutral chromatic group \(C_0(\mathcal G)\) [1001.4216]. A gain graph \(\Phi=(\Gamma,\phi)\) consists of an underlying graph \(\Gamma\) together with a gain mapping into a group \(\mathcal G\), with switching defined by \[ \phi^\eta(e_{ij}) := \eta(v_i)\,\phi(e_{ij})\,\eta(v_j)^{-1}. \] Weak neutral chromatic functions satisfy deletion–contraction on neutral links, \[ F(\Phi)=F(\Phi\setminus e)-F(\Phi/e), \] and vanish on neutral loops [1001.4216]. Two key gain-graph identities generalize the classical Chung–Graham expansions. If \(\Gamma_0\) is the neutral subgraph with neutral edge set \(E_0\), then \[ \Phi = \sum_{S\subseteq E_0} (-1)^{|S|} \bigl[(\Phi/S)\setminus E_0\bigr], \] which is the monomial-type expansion, and \[ \Phi = \sum_{\pi\in \Pi^*(\Gamma_0)} \bigl[(\Phi/\pi)\cup 1_{\mathcal G}K_\pi\bigr], \] which is the falling-factorial-type expansion [1001.4216]. Applying weak neutral chromatic invariants to these group identities produces expansions for the zero-free chromatic polynomial \(X_\Phi(q)\), the total chromatic polynomial \[ \widehat{X}_\Phi(q,z) = \sum_{S\subseteq E} (-1)^{|S|}q^{b(S)}z^{c(S)-b(S)}, \] and integral or modular chromatic functions [1001.4216]. The same framework yields explicit formulas for gain graphs corresponding to the Catalan, Shi, and Linial arrangements. For example, \[ X_{C_n}(q)=q(q-n-1)^{n-1}, \qquad X_{S_n}(q)=q(q-n)^{n-1}, \] and \[ X_{L_n}(q) = q\sum_{\pi\in\Pi_n}\prod_{i=2}^k (q-k-d_i(\pi)), \] with the Linial polynomial also equal to Athanasiadis’s formula \[ P_{L_n}(q)=\frac{q}{2^n}\sum_{j=0}^n \binom{n}{j}(q-j)^{n-1} \] [1001.4216]. In this setting, the Chung–Graham representation is not a hierarchy of properties but a representational calculus. It expresses a graph or gain graph as a canonical linear combination of contracted minors or neutral completions, thereby transferring combinatorial information into basis expansions of chromatic invariants. ## 4. Polynomial representations for permutations with bounded drop size In permutation enumeration, the Chung–Graham representation concerns the descent distribution on permutations with bounded maximum drop size. For \(\pi\in S_n\), the maximum drop size is \[ \maxdrop(\pi)=\max\{i-\pi_i:1\le i\le n\}, \] and \(A_{n,k}\) denotes the set of permutations of \([n]\) with maximum drop size at most \(k\). The corresponding descent polynomial is \[ A_{n,k}(y)=\sum_{\pi\in A_{n,k}} y^{\mathrm{des}(\pi)} =\sum_{d\ge 0} E_k(n,d)\,y^d \] [1306.5428]. Chung, Claesson, Dukes, and Graham introduced polynomials \(P_k(x)\) through a recurrence for \(A_{n,k}(y)\), and the paper "On Permutations with Bounded Drop Size" develops their coefficient structure [1306.5428]. The central formulas are \[ Q_k(x)=x^k P_k(x), \qquad R_{n,k}(x)=Q_k(x)(1+x+\cdots+x^k)^{n-k}. \] The coefficients of \(Q_k\) and \(R_{n,k}\) refine descent statistics jointly with the last value of the permutation: \[ Q_n(x)=\sum_{0\le i,j\le n} E(n+1,i;j+1)\,x^{(n+1)i+j}, \] and \[ R_{n,k}(x)=\sum_{0\le i\le n}\sum_{0\le j\le k} E_k(n+1,i;n+1-k+j)\,x^{(k+1)i+j} \] [1306.5428]. A principal result is the symmetry of \(R_{n,k}(x)\): \[ R_{n,k}(x)=x^{(n+1)k}R_{n,k}(1/x). \] The paper gives a bijective proof by constructing an involution \(\varphi\) on \(A_{n,k}\). For a permutation \(T\), let \(T\leftarrow i\) denote the append-and-standardize operation obtained by appending \(i\) and increasing each existing entry at least \(i\) by \(1\). If \(T\in T_k(n,i;j)\), where \[ T_k(n,i;j)=\{T\in A_{n,k}: \mathrm{des}(T)=i,\ T_n=n-k+j\}, \] then \(\varphi\) maps \(T_k(n,i;j)\) bijectively to \(T_k(n,i';j')\), where \[ i'=\left\lfloor \frac{(n+1)k-(k+1)i-j}{k+1}\right\rfloor \] and \[ j'=(n+1)k-(k+1)i-j-(k+1)i' \] [1306.5428]. This use of “Chung–Graham representation” is therefore a refined polynomial encoding of bounded-drop Eulerian statistics. It is structurally analogous to the chromatic and quasirandom usages in that a complex combinatorial class is encoded by canonical coefficients with strong symmetry or unimodality. The same paper also treats the type \(B\) analogue, introducing \(T_k(x)\) and proving Hyatt’s unimodality conjecture by embedding its coefficients into a symmetric unimodal auxiliary polynomial \(H_k(x)\) [1306.5428]. ## 5. Integer representations using even-indexed Fibonacci numbers A different and now prominent meaning of the Chung–Graham representation is an additive numeration system based on Fibonacci numbers. The paper "Chung-Graham and Zeckendorf representations" states the Chung–Graham theorem as follows: every non-negative integer \(n\) can be uniquely written as \[ n=\sum_{i\ge 0} a_i F_{i+2}, \] with \[ a_i\in\{0,1,2\}, \qquad a_i=0\ \text{if } i \text{ is odd}, \] and with the admissibility rule \[ \text{if } i