---
title: Generalized Mycielski Construction
url: https://www.emergentmind.com/topics/generalized-mycielski-construction
type: topic
---

# Generalized Mycielski Construction

The generalized Mycielski construction is a family of graph transformations that extends the classical Mycielskian by replacing the single clone layer with a stack of lift layers capped by an apex. In the now-standard undirected form, the generalized Mycielskian \(M_r(G)\) of a graph \(G\) has vertex set \(V(G)\times\{0,1,\dots,r-1\}\cup\{z\}\), contains a copy of \(G\) in layer \(0\), reproduces the adjacencies of \(G\) between consecutive layers, and joins the apex \(z\) to every vertex in the top layer. The case \(r=2\) is the classical Mycielskian, and \(M_r(K_2)\cong C_{2r+1}\). This layered construction has been studied for chromatic and topological lower bounds, zero-error information-theoretic parameters, linear-algebraic graph bounds, orientation parameters, immersion theory, and several nonclassical analogues [2312.09224][1710.01072][2507.09811].

## 1. Definition, special cases, and notational conventions

For a simple undirected graph \(G\), the generalized Mycielskian \(M_r(G)\) is defined by
\[
V(M_r(G))=V(G)\times\{0,1,\dots,r-1\}\cup\{z\},
\]
with edges
\[
\begin{aligned}
E(M_r(G))&=\big\{\{(v,0),(w,0)\}:\{v,w\}\in E(G)\big\}\\
&\quad\cup\big\{\{(v,i),(w,j)\}:\{v,w\}\in E(G)\ \text{and}\ |i-j|=1\big\}\\
&\quad\cup\big\{\{(v,r-1),z\}:v\in V(G)\big\}.
\end{aligned}
\]
Layer \(0\) induces a copy of \(G\); for each \(i=0,\dots,r-2\), the bipartite graph between layers \(i\) and \(i+1\) mirrors the adjacencies of \(G\); and the apex is adjacent exactly to the top layer. The classical Mycielskian is recovered at \(r=2\), so \(M_2(G)=M(G)\). A canonical special case is \(M_r(K_2)\cong C_{2r+1}\), which makes odd cycles the simplest generalized Mycielski graphs [2312.09224][2507.09811].

The classical \(r=2\) construction preserves clique number and raises chromatic number by one:
\[
\chi(M(G))=\chi(G)+1,\qquad \omega(M(G))=\omega(G).
\]
For general \(r\), the chromatic behavior is more delicate. Many graphs satisfy \(\chi(M_r(G))=\chi(G)+1\) for all \(r\), but there are exceptions; one explicit example is \(\chi(M_3(\overline{C_7}))=\chi(\overline{C_7})\) [2312.09224][1710.01072].

Notation is not uniform across the literature. Several recent papers use \(M_2(G)\) for the classical case, whereas work on dependent arcs and cover graphs uses \(M_m(G)\) with \(M_1(G)\) equal to the ordinary Mycielskian, and immersion-theoretic work writes \(\mu_m(G)\), where \(\mu_2(G)\) is classical and \(\mu_1(G)\) is \(G\) with one universal vertex added [1202.6461][2105.05724]. This indexing shift is purely conventional but important when comparing formulas.

## 2. Chromatic and topological role

The generalized construction is central in topological lower bounds for graph coloring. Müller and Stehlík consider the classes \(\mathcal M_k\) obtained by starting from \(K_2\) and iterating generalized Mycielski operations with arbitrary parameters \(r\). Stiebitz’s theorem states that if \(G\in\mathcal M_k\), then
\[
\chi(G)\ge k.
\]
Their proof deduces this from a version of Ky Fan’s combinatorial lemma, and they further show that Stiebitz’s theorem is equivalent to the Borsuk–Ulam theorem. In this sense, generalized Mycielski graphs form a discrete family that exactly captures a classical topological obstruction to low chromatic number [1710.01072].

A complementary formulation appears in the box-complex approach. Simons, Tardif, and Wehlau define classes \(K_k\) recursively by \(K_2=\{K_2\}\) and \(K_k=\{M_n(G):G\in K_{k-1},\, n\in\mathbb Z_{>0}\}\), and prove that for every graph \(H\),
\[
\mathrm{coind}(B(H))+2
\]
is the largest \(k\) for which some generalized Mycielski graph \(G\in K_k\) admits a homomorphism to \(H\). They thus replace Borsuk graphs by finite generalized Mycielski test graphs in the measurement of the coindex of the box complex. The same paper introduces a polynomial-time solvable linear system—the signature system—whose unsolvability implies \(\mathrm{coind}(B(H))+2\le 3\), while solvability implies \(\chi(H)\ge 4\) [1601.04642].

These results separate two phenomena that coincide for the classical Mycielskian but not for arbitrary \(r\). On the one hand, iterating generalized Mycielski steps from \(K_2\) still forces large chromatic number in the sense of Stiebitz’s theorem. On the other hand, for a fixed input \(G\), a single step \(M_r(G)\) with \(r>2\) need not increase \(\chi\). This distinction is fundamental in later work on capacity and spectral parameters, where \(r=2\) often behaves rigidly and \(r>2\) remains open.

## 3. Capacity, theta-type invariants, and asymptotic parameters

Csonka and Simonyi study the generalized construction from the distinguishability-graph viewpoint of zero-error information theory. Using the OR-product,
\[
V(F\cdot G)=V(F)\times V(G),\qquad
E(F\cdot G)=\big\{\{(f,g),(f',g')\}:\{f,f'\}\in E(F)\ \text{or}\ \{g,g'\}\in E(G)\big\},
\]
they define the nonlogarithmic Shannon OR-capacity
\[
C_{\mathrm{OR}}(G)=\lim_{t\to\infty}\sqrt[t]{\omega(G^t)}.
\]
Their main theorem for the standard Mycielski construction states that if \(G\) attains its OR-capacity at finite length, meaning
\[
C_{\mathrm{OR}}(G)=\sqrt[k]{\omega(G^k)}
\]
for some \(k\), then
\[
C_{\mathrm{OR}}(M(G))>C_{\mathrm{OR}}(G).
\]
The proof combines a large-clique construction for powers of \(M(K_n)\) with an embedding lemma \(M(G^t)\subseteq [M(G)]^t\). For \(G=K_2\), this recovers the exact equality \(C_{\mathrm{OR}}(C_5)=\sqrt 5\), since \(M(K_2)\cong C_5\). For generalized Mycielskians, the odd-cycle case remains tractable because \(M_r(K_2)\cong C_{2r+1}\), and Bohman–Holzman’s construction implies \(C_{\mathrm{OR}}(M_r(K_2))>2\) for all \(r\). By contrast, for \(n,r\ge 3\) the paper proves that the natural “lift-and-apex” strategy cannot produce the required clique structure, so a general capacity-increase theorem for \(M_r(G)\) remains open [2312.09224].

The same paper proves that the complementary Lovász theta number is completely determined by its value on the original graph under the classical Mycielski step. Writing \(\bar\vartheta(G)=\vartheta(\overline G)\), one has
\[
\bar\vartheta\!\left(M(G)\right)=\frac{4}{3}\,\bar\vartheta(G)\,
\cos\!\left(\frac{1}{3}\arccos\!\Big(1-\frac{27}{4\,\bar\vartheta(G)}+\frac{27}{16\,\bar\vartheta(G)^2}\Big)\right)
-\frac{1}{3}\,\bar\vartheta(G)+1.
\]
This is proved twice: first through a lifted strict vector coloring, which yields a cubic equation for the new value, and second through Lovász’s spectral characterization. Special cases include \(\bar\vartheta(M(K_2))=\sqrt5\) and \(\bar\vartheta(M(K_3))=4\cos(2\pi/9)\) [2312.09224].

The fractional chromatic number exhibits an analogous value-determination phenomenon. Larsen–Propp–Ullman showed for \(r=2\) that
\[
\chi_f(M(G))=\chi_f(G)+\frac{1}{\chi_f(G)},
\]
and Tardif generalized this to all \(r\):
\[
\chi_f(M_r(G))=\chi_f(G)+\frac{1}{\sum_{k=0}^{r-1}(\chi_f(G)-1)^k}.
\]
Tamura proves that the complementary fractional Haemers bound satisfies the same Tardif-type law whenever it equals the clique number:
\[
\bar{\mathcal H}_f(M_r(G);\mathbb F)
=
\bar{\mathcal H}_f(G;\mathbb F)+
\frac{1}{\sum_{k=0}^{r-1}(\bar{\mathcal H}_f(G;\mathbb F)-1)^k}
\quad
\text{whenever }
\bar{\mathcal H}_f(G;\mathbb F)=\omega(G).
\]
More generally,
\[
\bar{\mathcal H}_f(M_r(G);\mathbb F)\le
\bar{\mathcal H}_f(G;\mathbb F)+
\frac{1}{\sum_{k=0}^{r-1}(\bar{\mathcal H}_f(G;\mathbb F)-1)^k},
\]
with equality proved for cliques and for every graph satisfying the clique-equality hypothesis. No counterexample is known to the conjecture that this exact formula holds for all graphs and fields [2507.09811].

These formulas place generalized Mycielskians inside the asymptotic-spectrum program. Csonka and Simonyi explicitly frame \(\chi_f\), \(\bar\vartheta\), and Mycielski-determined parameters through classes such as \(\mathcal{AS}\), \(\mathcal C_{\mathrm{Myc}}\), and \(\mathcal C_{\mathrm{genMyc}}\), and ask whether broader systematic relations hold between asymptotic-spectrum parameters and Mycielski determination [2312.09224].

## 4. Other graph invariants and structural consequences

The generalized construction has also been used to study invariants far removed from chromatic theory. For boxicity, Adiga, Chandran, and Sivadasan prove the monotonicity
\[
\operatorname{box}(M(G))\ge \operatorname{box}(G),
\]
and, more precisely, if \(G\) has \(l\) universal vertices, then
\[
\operatorname{box}(M(G))\ge \operatorname{box}(G)+\left\lceil\frac l2\right\rceil,
\]
while
\[
\operatorname{box}(M(G))\le \theta(\overline G)+\left\lceil\frac l2\right\rceil+1,
\]
with the stronger upper bound
\[
\operatorname{box}(M(G))\le \theta(\overline G)+\left\lceil\frac l2\right\rceil
\]
when \(l\) is zero or odd. They compute
\[
\operatorname{box}(M(K_n))=
\begin{cases}
\left\lceil n/2\right\rceil,& n\ \text{odd},\\[2mm]
\left\lceil n/2\right\rceil+1,& n\ \text{even},
\end{cases}
\]
and extend the universal-vertex lower bound to every \(M_r(G)\). They also show that for bipartite \(G\),
\[
\operatorname{box}(M_r(G))\le \operatorname{box}(M_2(G))+2,
\]
and for arbitrary \(G\) and \(r\ge 3\),
\[
\operatorname{box}(M_r(G))\le \operatorname{box}(M_3(G))+1.
\]
This reduces much of the generalized problem to the cases \(r=2\) and \(r=3\) [1308.2368].

In orientation theory, Wang, Chang, and Tian examine the minimum number \(m(G)\) of dependent arcs over all acyclic orientations of \(G\), where an arc is dependent if reversing it creates a directed cycle. In their indexing, the generalized Mycielski graph \(M_m(G)\) has classical case \(M_1(G)\). They prove that
\[
m(M_m(G))\ge 1 \quad\Longleftrightarrow\quad G\ \text{is not bipartite},
\]
equivalently, \(M_m(G)\) is a cover graph precisely when \(G\) is bipartite. They also give sufficient conditions for \(m(M_m(G))=1\), prove additive lower bounds such as \(d_{\min}(M_m(G))\ge d_{\min}(G)+2\) for triangle-free \(G\) with \(d_{\min}(G)\ge1\), and obtain parallel results for the edge-deletion parameter \(c(G)\) leading to a cover graph. Odd cycles satisfy \(m(M_m(C_{2t+1}))=1\) and \(c(M_m(C_{2t+1}))=1\) [1202.6461].

For clique immersions, Burger and Andreae prove that generalized Mycielskians increase immersion number by at least one:
\[
\operatorname{im}(\mu_m(G))\ge \operatorname{im}(G)+1.
\]
Their proof uses the “distinct neighbor property” of \(K_t\)-immersions to route edge-disjoint paths from the base layer to the root. The lower bound is best possible for \(m=1\) and for \(\mu_2(K_t)\), where \(\operatorname{im}(\mu_2(K_t))=t+1\). In other families it is not sharp: for \(n\ge5\) and \(m\ge3\), \(\operatorname{im}(\mu_m(P_n))=5\), while for cycles \(C_n\) with \(n\ge5\), \(\operatorname{im}(\mu_m(C_n))=4\) at \(m=1\) and \(5\) for \(m\ge2\) [2105.05724].

## 5. Directed and quantum generalizations

The directed version replaces graphs by digraphs and preserves orientations between adjacent layers. For a digraph \(D\), the generalized directed Mycielskian \(M_r(D)\) has the same layered vertex set \(V(D)\times\{0,\dots,r-1\}\cup\{z\}\), with arcs inside layer \(0\) copied from \(D\), arcs between consecutive layers following the original orientations in both relative directions, and arcs from the apex to the top layer. In the \(r=2\) case,
\[
V(M(D))=V(D)\times\{0,1\}\cup\{z\},
\]
and the apex has outward arcs to layer \(1\). Csonka and Simonyi prove the directed analogue of the finite-attainment theorem: if a digraph \(D\) attains its Sperner capacity at finite length, then
\[
C_{\mathrm{Sp}}(M(D))>C_{\mathrm{Sp}}(D).
\]
The unique orientation of \(C_5\) is \(M(T_2)\), and its Sperner capacity is \(\sqrt5\), whereas other orientations of \(C_5\) have Sperner capacity \(2\). For generalized directed odd cycles \(M_r(T_2)\), the inequality \(C_{\mathrm{Sp}}>2\) is known for \(r=2\) and \(r=3\), and the extension to \(r\ge4\) remains open [2312.09224].

A noncommutative extension appears in quantum graph theory. Bochniak, Chełstowski, Kasprzak, and Sołtan define the quantum Mycielskian \(p_{r-1}(G)\) of a quantum graph \(G=(V_G,\varphi_G,A_G)\) by
\[
C(p_{r-1}(G))=\mathbb C\oplus \bigoplus_{k=1}^r C(G),
\]
equipped with a block adjacency operator that couples the master summand to the top layer and successive layers via the original adjacency \(A_G\). This reproduces the classical construction on commutative graphs. They prove that quantum isomorphism is preserved under \(p_{r-1}\), that \(QAut(G)\) lifts to an action on \(p_{r-1}(G)\), and that if \(G\) has no quantum twin vertices then, for every partition of unity \(\mathcal P\),
\[
QAut(p(G))_{\mathcal P^{(1)}} = QAut(G)_{\mathcal P}.
\]
Under the same no-twin hypothesis, the quantum distinguishing number satisfies
\[
D(p(G))\le D(G)+1.
\]
Their paper also notes earlier results showing that quantum clique number is invariant under \(p_{r-1}\), while quantum chromatic number can increase by at most \(1\) under the basic step \(p=p_1\) [2502.10521].

These extensions preserve the formal idea of a layered lift with an apex, but the preserved parameter depends strongly on context. In directed capacity theory the key object is the transitive clique number in OR-powers; in quantum graph theory it is symmetry under compact quantum groups. The construction is therefore better regarded as a schema—copy, lift, and cap—than as a single invariant-preserving mechanism.

## 6. Related Mycielski-like constructions, applications, and open problems

Several constructions are closely related to generalized Mycielskians without being identical to \(M_r(G)\). The “great shadow” \(S(G)\) duplicates each vertex \(v\) by a shadow \(v'\) adjacent to \(v\) and to every neighbor of \(v\). Unlike the classical Mycielskian, this immediately creates triangles whenever \(G\) has an edge. DeJong, Lee, and Shields prove the sharp planarity criterion
\[
S(G)\ \text{is planar}\quad\Longleftrightarrow\quad G\ \text{is a bipartite cactus graph}.
\]
They motivate \(S(G)\) by diode-and-switch keyboard circuits, where single-sided PCB routability is equivalent to planarity of the great shadow. In this variant, the Mycielski-type duplication mechanism is used not for chromatic amplification but for planar embeddability and circuit layout [2505.09534].

The name “Mycielski” also appears in constructions that are not graph lifts. In topological MV-algebras, He and Li study the Hartman–Mycielski object \(A^\bullet\), the MV-algebra of step functions \(J=[0,1)\to A\) with the HM-topology. For every Hausdorff topological MV-algebra \(A\), the canonical embedding \(i_A:A\to A^\bullet\) identifies \(A\) with a closed subalgebra of a pathwise connected, locally pathwise connected topological MV-algebra. Bounded continuous pseudometrics and bounded continuous real-valued functions extend to \(A^\bullet\) by integration, and the functor \(A\mapsto A^\bullet\) preserves open surjections and quotients [2606.08541]. In descriptive set theory, Mycielski-type theorems have been generalized from perfect squares to rectangles \([P]\times[M]\) or \([P]\times[P]\) whose sides are bodies of Miller or uniformly perfect trees, with sharp positive and negative results depending on whether the ambient largeness notion is category or measure [1905.09069]. These uses share the name and a “largeness-by-structured-lift” motif, but they are distinct from the graph-theoretic \(M_r(G)\).

Within graph theory, several problems remain open. For \(r>2\), a capacity-increase theorem comparable to the \(r=2\) Shannon-capacity result is not known in general, and extending the explicit \(\bar\vartheta\)-formula from \(M(G)\) to \(M_r(G)\) is also open [2312.09224]. Tamura’s exact formula for the complementary fractional Haemers bound is proved only under the hypothesis \(\bar{\mathcal H}_f(G;\mathbb F)=\omega(G)\), although no counterexample to the general conjecture is known [2507.09811]. In quantum symmetry theory, the status of quantum twin vertices for \(K_2\square K_5\) and \(C_4\square C_3\) is left open [2502.10521]. In immersion theory, Burger and Andreae conjecture that for \(m\ge3\),
\[
\operatorname{im}(\mu_m(K_{m+1}))=2m+1
\]
[2105.05724]. Csonka and Simonyi also ask whether
\[
\lim_{t\to\infty} C_{\mathrm{OR}}(M^t(K_2))
\]
is finite, a question they identify as equivalent to a well-known Ramsey-type problem [2312.09224].

Across these variants and open directions, the generalized Mycielski construction remains a precise mechanism for transferring local adjacency data into a higher-layer object whose global parameters are often more rigid, sometimes universal, and frequently still only partially understood.

Source: https://www.emergentmind.com/topics/generalized-mycielski-construction