---
title: 'Generalized Johnson Graphs: Structure & Metrics'
url: https://www.emergentmind.com/topics/generalized-johnson-graphs
type: topic
---

# Generalized Johnson Graphs: Structure & Metrics

to=arxiv_search.search เติมเงินไทยฟรี 70 200
{"query":"generalized Johnson graph 2304.02864 girth odd girth distance diameter", "max_results": 10}
to=arxiv_search.search  彩神争霸代理 70 200
{"query":"The girth, odd girth, distance function, and diameter of generalized Johnson graphs", "max_results": 5}
to=arxiv_search.search ,一本道 70 200
{"query":"merged Johnson graphs distinguishing number 1106.1007", "max_results": 5}
to=arxiv_search.search 】【。】【”】【70 200
{"query":"zero blocking numbers generalized Johnson graphs 2508.02125", "max_results": 5}
to=arxiv_search.search 】【。】【”】【70 200
{"query":"On the diameter and zero forcing number of some graph classes in the Johnson Grassmann and Hamming association scheme generalized Johnson graphs 2302.07757", "max_results": 5}
to=arxiv_search.search 】【。】【”】【70 200
{"query":"Properties of Full-Flag Johnson Graphs 1301.1131", "max_results": 5}
to=arxiv_search.search 】【。】【”】【70 200
{"query":"On the Connectedness and Diameter of a Geometric Johnson Graph 1202.3455", "max_results": 5}
Generalized Johnson graphs are intersection graphs on uniform set systems in which vertices are \(k\)-subsets of an \(n\)-element ground set and adjacency is controlled by prescribed intersection cardinalities. In the exact-intersection form, the graph is denoted \(J(v,k,i)\), \(J(n,k,a)\), or \(GJ(n,k,l)\), with two vertices adjacent exactly when their intersection has size \(i\), \(a\), or \(l\), respectively. This framework contains the classical Johnson graph at parameter \(k-1\) and the Kneser graph at parameter \(0\), and it extends naturally to multi-intersection graphs \(J_S(n,k)\), geometric island graphs \(IJ(P,k,l)\), and full-flag analogues on permutations [2304.02864] [2302.07757] [1202.3455] [1301.1131].

## 1. Definitions, notation, and principal families

The basic exact-intersection model fixes integers with \(v>k>i\) or \(n\ge k>a\ge 0\), takes vertex set \(\binom{[n]}{k}\), and joins distinct \(k\)-subsets \(A,B\) when \(|A\cap B|=i\) or \(|A\cap B|=a\). In this sense, the notation varies across the literature, but the underlying object is the same single-intersection graph. The most common specializations are the Johnson graph, obtained when the required intersection is \(k-1\), and the Kneser graph, obtained when the required intersection is \(0\) [2304.02864] [2508.02125] [1202.3455].

A broader form replaces a single allowed intersection size by a set \(S\subseteq\{0,1,\dots,k-1\}\). The graph \(J_S(n,k)\) has the same vertex set \(\binom{[n]}{k}\) and adjacency condition \(|A\cap B|\in S\). This includes the usual Johnson graph when \(S=\{k-1\}\), the Kneser graph when \(S=\{0\}\), and the “generalized Kneser graph” \(J_{\{0,1,\dots,t\}}(n,k)\) when all intersections up to \(t\) are allowed [2302.07757].

A recurrent source of terminological ambiguity is that several non-equivalent objects are also described as generalized Johnson graphs. The geometric model \(IJ(P,k,l)\) replaces all \(k\)-subsets by \(k\)-islands of a planar point set \(P\), while Full-Flag Johnson graphs \(FJ(n,k)\) replace subsets by full flags, equivalently by permutations of \([n]\). These are genuine extensions of the Johnson paradigm, but they are structurally different from the exact-intersection graphs on \(\binom{[n]}{k}\) [1202.3455] [1301.1131].

| Family | Vertex set | Adjacency rule |
|---|---|---|
| \(J(v,k,i)\), \(J(n,k,a)\), \(GJ(n,k,l)\) | \(k\)-subsets of \([n]\) | \(|A\cap B|=i,a,l\) |
| \(J_S(n,k)\) | \(k\)-subsets of \([n]\) | \(|A\cap B|\in S\) |
| \(IJ(P,k,l)\) | \(k\)-islands of \(P\) | \(|I\cap J|=l\) |
| \(FJ(n,k)\) | permutations of \([n]\) | exactly \(k\) initial segments differ |

Within the exact-intersection theory, it is standard to assume \(v\ge 2k\) and \((v,k,i)\neq(2k,k,0)\), and to set
\[
\Delta=v-2k+2i.
\]
The parameter \(\Delta\) governs how much an intersection can grow or shrink in one step, and it organizes the formulas for girth, odd girth, distance, and diameter [2304.02864].

## 2. Metric structure of \(J(v,k,i)\)

For \(X=J(v,k,i)\) under the assumptions \(v>k>i\), \(v\ge 2k\), and \((v,k,i)\neq(2k,k,0)\), the girth is
\[
g(X)=
\begin{cases}
3, &v\ge 3\,(k-i),\\
4, &v<3\,(k-i)\quad\text{and}\quad (v,k,i)\neq(2k+1,k,0),\\
5, &(v,k,i)=(5,2,0),\\
6, &(v,k,i)=(2k+1,k,0)\;\text{with }k>2.
\end{cases}
\]
The triangle threshold is exact: a \(3\)-cycle exists precisely when adjacent \(k\)-subsets admit a common neighbor, which reduces to the inequality \(v\ge 3(k-i)\). Outside the two exceptional odd-graph cases, a \(4\)-cycle can be built explicitly; the remaining exceptions are the classical odd graphs \(J(2k+1,k,0)\), with girth \(6\), and the small Kneser case \(J(5,2,0)\), with girth \(5\) [2304.02864].

The odd girth has a uniform closed form:
\[
og(X)=2\Bigl\lceil\frac{k-i}{\Delta}\Bigr\rceil+1.
\]
The lower bound comes from analyzing an odd cycle of length \(2r+1\) through a vertex at distance \(r\) from two adjacent vertices, and the matching upper bound is obtained by constructing a closed walk of that length with carefully prescribed pairwise intersections. This shows that the odd-girth problem is completely controlled by the ratio \((k-i)/\Delta\) [2304.02864].

The distance between two vertices depends only on \(x=|A\cap B|\). The exact distance formula is
\[
\mathrm{dist}_X(A,B)=
\begin{cases}
3, &x<\min\{\,i,\;k-\Delta\},\\
\left\lceil\dfrac{k-x}{\,k-i\,}\right\rceil, &k-\Delta\le x<i,\\
\displaystyle
\min\!\Bigl\{
2\Bigl\lceil\frac{k-x}{\Delta}\Bigr\rceil,\;
2\Bigl\lceil\frac{x-i}{\Delta}\Bigr\rceil+1
\Bigr\}, &x\ge i.
\end{cases}
\]
For \(x<i\), the issue is whether \(A\) and \(B\) have a common neighbor; if not, paths are built by replacing up to \(k-x\) elements in steps of size \(k-i\). For \(x\ge i\), there is a two-way competition between decreasing the intersection from \(x\) to \(i\) in \(\Delta\)-steps and increasing it from \(x\) toward \(k\) in \(\Delta\)-steps, giving the two quantities inside the minimum [2304.02864].

The diameter is then
\[
\mathrm{diam}(X)=
\begin{cases}
\left\lceil\dfrac{k-i-1}{\Delta}\right\rceil+1,
&v<3(k-i)-1\quad\text{or}\quad i=0,\\
3,
&3(k-i)-1\le v<3k-2i\quad\text{and}\quad i\neq 0,\\
\left\lceil\dfrac{k}{\,k-i\,}\right\rceil,
&v\ge 3k-2i\quad\text{and}\quad i\neq 0.
\end{cases}
\]
This recovers the known Kneser-graph diameter \(\left\lceil(k-1)/(v-2k)\right\rceil+1\) when \(i=0\), and the Johnson-graph diameter \(k\) when \(i=k-1\). A plausible implication is that \(\Delta\) functions as the effective local step size for the entire metric theory of exact-intersection generalized Johnson graphs [2304.02864].

## 3. Connectivity and diameter beyond the exact metric formulas

For the classical exact-intersection graph \(GJ(n,k,l)\), connectedness is governed by a simple threshold:
\[
GJ(n,k,l)\text{ is connected if and only if }n\ge 2k-l.
\]
If \(n<2k-l\), there exist \(k\)-sets with intersection smaller than \(l\), and no walk with successive intersections exactly \(l\) can join them. If \(n\ge 2k-l\), an exchange argument replaces elements of \(A\setminus B\) by elements of \(B\setminus A\) in batches of size \(k-l\), always maintaining overlap \(l\) [1202.3455].

Under connectedness, the classical diameter admits the bound
\[
\mathrm{dist}_{GJ}(A,B)\le
\left\lceil\frac{k-|A\cap B|}{k-l}\right\rceil
\le
\left\lceil\frac{k}{k-l}\right\rceil,
\]
hence
\[
\mathrm{diam}(GJ(n,k,l))\le \left\lceil\frac{k}{k-l}\right\rceil.
\]
For \(1\le l<k\), one has
\[
\mathrm{diam}(GJ(n,k,l))=\Theta\!\bigl(\tfrac{k}{k-l}\bigr),
\]
while the special cases recover \(\mathrm{diam}(J(n,k))=k\) and \(\mathrm{diam}(K(n,k))=2\) for \(n\ge 2k\) [1202.3455].

The geometric variant \(IJ(P,k,l)\) substantially changes the picture. Here \(P\) is a planar point set in general position, a \(k\)-island is a \(k\)-subset realized as \(P\cap C\) for some convex set \(C\), and two islands are adjacent when their intersection has size \(l\). When \(P\) is in convex position, every \(k\)-subset is an island, so \(IJ(P,k,l)\cong GJ(n,k,l)\). For arbitrary \(P\), however, connectedness may require the stronger condition
\[
n>(k-l)(k-l+1)+k.
\]
Under that hypothesis,
\[
\mathrm{diam}(IJ(P,k,l))=O\!\Bigl(\frac{n}{k-l}+(k-l)\Bigr).
\]
If \(l\le k/2\) and
\[
n\ge 2\bigl((k-l)(k-l+1)+k\bigr),
\]
then the upper bound improves to
\[
\mathrm{diam}(IJ(P,k,l))=O(\log n)+(k-l).
\]
At the same time, Horton-set constructions yield the lower bound
\[
\mathrm{diam}(IJ(P,k,l))=\Omega\!\Bigl(\frac{\log n-\log k}{\log(k-l)}\Bigr),\qquad l\ge 2.
\]
These results show that the geometric restriction from all \(k\)-subsets to convexly realizable \(k\)-subsets can raise both connectivity thresholds and diameter growth rates [1202.3455].

## 4. Symmetry, automorphisms, and distinguishing colorings

The notation \(\mathrm{MJ}(n,k,s)\) is used for a graph with vertex set \(\binom{[n]}{k}\) and adjacency
\[
|S\cap T|=k-s.
\]
Thus \(\mathrm{MJ}(n,k,1)=J(n,k)\) and \(\mathrm{MJ}(n,k,k)=K(n,k)\). In this formulation, merged Johnson graphs sit on the same vertex set as generalized Johnson graphs and recover the classical Johnson and Kneser graphs as extremal instances [1106.1007].

The principal invariant studied in this setting is the distinguishing number \(\Dist(G)\), the least number of vertex-colors for which the only color-preserving automorphism is the identity. A complete piecewise determination is given for \(\Dist(\mathrm{MJ}(n,k,s))\). The proof strategy combines orbital-graph analysis of automorphism groups, determining sets, asymmetric induced subgraphs, and in one family a probabilistic coloring argument. The automorphism group is \(S_n\) in most cases, with exceptional families involving \(S_{n+1}\), \(S_2\wr S_n\), or \(S_2\times S_n\) [1106.1007].

The generic outcome is \(2\)-distinguishability. The key mechanism is that one can exhibit a small induced subgraph \(H\subset J\) which is asymmetric and yet is a determining set for \(J\); then the Albertson–Collins theorem yields \(\Dist(J)\in\{1,2\}\), and since the graph is not asymmetric, the value is \(2\). Exceptional balanced cases require more colors. In particular, when \(n=2m\) and \(s=m\), the graph is a perfect matching of size \(m\), and a counting argument on equipartitions yields
\[
\Dist\bigl(\mathrm{MJ}(2m,m,m)\bigr)=3.
\]
The examples \(\mathrm{MJ}(5,2,1)=J(5,2)\) with distinguishing number \(3\) and \(\mathrm{MJ}(6,3,1)=J(6,3)\) with distinguishing number \(2\) illustrate the difference between small exceptional parameters and the generic regime [1106.1007].

This line of work places generalized Johnson-type graphs within algebraic graph theory as highly symmetric but not uniformly rigid objects. A plausible implication is that the symmetry-breaking problem is controlled less by local degree than by exceptional automorphism enlargements at balanced parameters.

## 5. Zero forcing, failed forcing, and zero blocking

Zero forcing on a graph \(G\) begins with a black-white coloring of \(V(G)\) and applies the rule: if a black vertex has exactly one white neighbor, that neighbor becomes black. A zero forcing set is an initial black set that eventually turns every vertex black; its minimum size is the zero forcing number \(Z(G)\). A failed zero forcing set is an initial black set for which the process stalls before all vertices become black; the maximum size of such a set is \(F(G)\). The complement of a failed zero forcing set is a zero blocking set, and its minimum size is the zero blocking number \(B(G)\), with
\[
F(G)+B(G)=|V(G)|.
\]
These parameters have been studied for generalized Johnson graphs in both the multi-intersection and single-intersection settings [2508.02125] [2302.07757].

For \(J_S(n,k)\) with \(s=\min S\) and \(n\ge 2k-s\), an explicit construction gives a connected zero forcing set of size
\[
\binom{n}{k}-\binom{n-2(k-s)}{s},
\]
so
\[
Z\bigl(J_S(n,k)\bigr),\; Z_t,\; Z_c
\le
\binom{n}{k}-\binom{n-2(k-s)}{s}.
\]
When \(S=\{s,s+1,\dots,k-1\}\), equality holds throughout. In the “\(\le t\)” case \(S=\{0,1,\dots,t\}\) with \(n\ge 2k+1\), one obtains the exact value
\[
Z\bigl(J_{\{0,\dots,t\}}(n,k)\bigr)
=
\binom{n}{k}-\binom{n-2(k-t)}{k-t}.
\]
Setting \(t=k-1\) recovers the known formula for the Johnson graph,
\[
Z(J(n,k))=\binom{n}{k}-\binom{n-2}{k-1}.
\]
The proofs use a Bollobás-type intersection lemma and Z–Grundy bounds [2302.07757].

For the exact-intersection graph \(J(n,k,a)\), the zero blocking number behaves differently. If \(n\le 2k-a-1\), then the graph is edgeless and \(B=1\). If \((n,k,a)=(k+1,k,a)\) with \(a=k-2\), or \((n,k,a)=(2k,k,0)\), or \(n=2k=4a\), then the graph has twins and \(B=2\). In all other cases with \(n\ge 2k-a\) and \(k\ge 2\), the graph has no isolated vertices and no twins, so \(B\ge 3\) [2508.02125].

There is also a universal upper bound:
\[
B\bigl(J(n,k,a)\bigr)\le \max\{a,k-a\}+2,\qquad 1\le a\le k-2.
\]
For sufficiently large \(n\), this is exact. Writing \(\Delta=\max\{a,k-a\}\), one has
\[
B\bigl(J(n,k,a)\bigr)=\max\{a,k-a\}+2
\quad\text{as soon as } n\gg k,a.
\]
More explicitly, if \(k-2\ge a\ge k/2\) and
\[
n\ge \Bigl(a+\tfrac52\Bigr)(k-a+2)-6,
\]
then
\[
B\bigl(J(n,k,a)\bigr)=a+2,
\]
while if \(k-2\ge a<k/2\) and
\[
n\ge k(k-a+2)-a,
\]
then
\[
B\bigl(J(n,k,a)\bigr)=(k-a)+2.
\]
The significance of these results is not merely combinatorial: zero forcing connects to maximum nullity and minimum rank, while zero blocking measures how effectively one can defeat the forcing process [2508.02125].

## 6. Full-flag and \(q\)-analogues

Full-Flag Johnson graphs \(FJ(n,k)\) replace \(k\)-subsets by full flags \(U_1\subset U_2\subset\cdots\subset U_n\) with \(|U_i|=i\), equivalently by permutations \(u=(u_1,\dots,u_n)\) of \([n]\) via \(U_i=\{u_1,\dots,u_i\}\). Two vertices \(u,v\) are adjacent precisely when exactly \(k\) of the initial-segment sets differ; equivalently,
\[
|\{i:u(i)=v(i)\}|=n-k.
\]
The special cases are \(FJ(n,0)\), the isolated graph, and \(FJ(n,1)\), exactly the permutahedron [1301.1131].

These graphs admit a Cayley-graph description:
\[
FJ(n,k)\cong \mathrm{Cay}(S_n,T_{n,k}),
\]
where \(T_{n,k}\) consists of all permutations whose permutation matrix is \((n-k)\)-block diagonal, equivalently decomposes into exactly \(n-k\) irreducible blocks along the diagonal. For \(k=1\), the generating set is the neighboring transpositions; for \(k=n-1\), it is all irreducible permutations, and the diameter is \(2\). The adjacency matrix also has a recursive block structure: after ordering vertices by inserting \(n+1\) in each possible position, \(A(FJ(n+1,k))\) decomposes into \((n+1)^2\) blocks, with zero blocks when \(|i-j|>k\), corner diagonal blocks equal to \(A_n,k\), and \(|i-j|=1\) blocks equal to \(A_n,k-1\) [1301.1131].

For the permutahedron \(FJ(n,1)\), this block structure yields an \(n\times n\) regularity matrix \(M\) whose eigenvalues form a subset of those of \(A(FJ(n,1))\). The example
\[
\mathrm{Spec}\,FJ(4,1)=\{3,1+\sqrt2,\sqrt3,1,-1+\sqrt2,1-\sqrt2,-1,-\sqrt3,-1-\sqrt2,-3\}
\]
and
\[
\mathrm{Spec}\,M=\{3,1+\sqrt2,1,1-\sqrt2\}
\]
illustrates the partial spectral information accessible from the recursive construction [1301.1131].

A different extension moves from sets to subspaces. The generalized Grassmann graph \(J_{q,S}(n,k)\) has as vertices the \(k\)-dimensional subspaces of \(\mathbb F_q^n\), with adjacency determined by \(\dim(U\cap W)\in S\). If \(s=\min S\) and \(n\ge 2k\), then
\[
\mathrm{diam}\,J_{q,S}(n,k)=
\begin{cases}
2,& s=0,\\
\left\lceil\tfrac{k}{k-s}\right\rceil,&1\le s\le k-1,
\end{cases}
\]
and if \(S\neq\emptyset\), then \(\mathrm{girth}=3\). For \(S=\{0,1,\dots,t\}\) and \(n\ge 2k+1\), zero-forcing formulas parallel the set-system case with Gaussian binomial coefficients replacing ordinary binomial coefficients. This places generalized Johnson-type graphs inside a wider Johnson–Grassmann–Hamming association-scheme framework [2302.07757].

Across these variants, the common principle is that adjacency is defined by controlled overlap—cardinality for sets, dimension for subspaces, convex realizability for islands, and rank-by-rank agreement for full flags. This suggests that generalized Johnson graph theory is best viewed not as a single graph family, but as a cluster of intersection-type and overlap-type models whose algebraic, metric, and propagation properties can often be studied by the same counting and exchange methods.

Source: https://www.emergentmind.com/topics/generalized-johnson-graphs