Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated

Further restrictions on two-cover diameter graphs


All multiplicities count unordered pairs. These results extend the star-diameter theorem, but do not settle either unrestricted question.

A bound for a two-vertex diameter cover

Let D>rD>r be the two largest distances. Suppose the diameter graph has vertex-cover number exactly two, and choose a cover {a,c}\{a,c\}. Put

B=ND(a)∖{c},E=ND(c)∖{a},η=∣B∩E∣.B=N_D(a)\setminus\{c\},\quad E=N_D(c)\setminus\{a\},\quad \eta=|B\cap E|.

Then η≤1\eta\le1 and

mA(D)+mA(r)≤n+8+2η.(1)m_A(D)+m_A(r)\le n+8+2\eta. \tag{1}

In particular, rr is rare if mA(D)≥10m_A(D)\ge10, or if η=0\eta=0 and mA(D)≥8m_A(D)\ge8.

Normalize r=1r=1, and let C=A∖{a,c}C=A\setminus\{a,c\}, of diameter at most one. Both cover points have diameter neighbors in CC, since the graph is not a star. Write Na,NcN_a,N_c for their unit neighbors in CC, and put h=∣Na∩Nc∣h=|N_a\cap N_c|, ε=1∣a−c∣=1\varepsilon=\mathbf1_{|a-c|=1}. If zz is the number of globally unit-inactive vertices, then

mA(1)≤n+2+h+ε−z.(2)m_A(1)\le n+2+h+\varepsilon-z. \tag{2}

Indeed, NaN_a lies on an arc of angular width at most π/3\pi/3: its points are within one of a fixed diameter neighbor of aa, initially restricting them to an arc of width less than 2π/32\pi/3, and its extreme members are at distance at most one. The unit-circle arc lemma makes every nonextreme member isolated in the unit graph on CC. The same applies to NcN_c. If t=∣Na∪Nc∣t=|N_a\cup N_c|, at least t−4t-4 of these points are isolated on CC, in addition to globally inactive vertices of CC. Apply the diameter bound after removing these isolates and then add the t+h+εt+h+\varepsilon unit edges touching the cover. This proves (2) when both cover vertices are active. An inactive cover vertex removes two possible extreme positions from this count, so (2) remains valid with all inactive vertices included in zz.

We have h≤1h\le1. Otherwise two common neighbors may be written x=(−u,0)x=(-u,0), y=(u,0)y=(u,0) and the cover points as a=(0,H)a=(0,H), c=(0,−H)c=(0,-H), with H2=1−u2H^2=1-u^2. Every point w∈Aw\in A lies in both unit disks about x,yx,y, and hence

∣w∣2+2u∣w1∣≤H2.|w|^2+2u|w_1|\le H^2.

Only a,ca,c lie on the boundary of the enclosing radius-HH disk. Thus acac is the unique diameter, contrary to the cover number.

There is at most one common diameter neighbor. Two such neighbors b,d∈Cb,d\in C would give ∣a−c∣2=4D2−∣b−d∣2>D2|a-c|^2=4D^2-|b-d|^2>D^2. Each of B,EB,E lies on a radius-DD arc of angular width less than π/3\pi/3. The radius-DD arc inequality in star_diameter makes every interior member of BB unit-inactive in CC; its only possible unit neighbor in AA is cc. If η=0\eta=0, the same inequality applied to cc rules out that neighbor. If η=1\eta=1, the common diameter neighbor is an extreme of the arc (otherwise its distances to the two extremes give the same contradiction). The radius-DD circle about aa meets the unit disk about cc in an arc of angular width less than π\pi. One extreme of BB lies outside that disk and the other lies inside or on it. Consequently at most one interior member of BB lies on the unit circle about cc. The identical argument applies to EE.

The interiors of the two arcs are disjoint, so, writing δ=1∣a−c∣=D\delta=\mathbf1_{|a-c|=D},

z≥∣B∣+∣E∣−4−2η=mA(D)−δ−4−2η.z\ge |B|+|E|-4-2\eta=m_A(D)-\delta-4-2\eta.

Now (2), h≤1h\le1, and δ+ε≤1\delta+\varepsilon\le1 prove (1).

A regular even polygon cannot acquire a noncentral point

A regular even mm-gon, m≥4m\ge4, admits no new noncentral point whose distances to all its vertices belong to its chord palette. Its center is admissible exactly when 6∣m6\mid m.

Normalize its circumradius to one and write m=2sm=2s. A noncentral point pp has at most two vertices at each prescribed distance, so all ss chord lengths must occur exactly twice among its mm pinned distances. Since ∑j=1scos⁡(πj/s)=−1\sum_{j=1}^s\cos(\pi j/s)=-1, averaging squares gives

∣p∣2=1+2/s=1+4/m.|p|^2=1+2/s=1+4/m.

There is a vertex within angle π/m\pi/m of the direction −p-p, and its squared distance from pp is at least

1+∣p∣2+2∣p∣cos⁡(π/m)>4,1+|p|^2+2|p|\cos(\pi/m)>4,

using cos⁡(π/m)≥1−2/m\cos(\pi/m)\ge1-2/m and ∣p∣>1|p|>1. This exceeds the polygon diameter. For the center, admissibility is exactly 2sin⁡(πj/m)=12\sin(\pi j/m)=1, or 6∣m6\mid m.

Consequently, if nn is even, the only diameter pairs are two disjoint pairs, and every point is second-distance-active, then R(A)≥2R(A)\ge2. Under R(A)=1R(A)=1, the pair budget gives q(A)≤n/2−1q(A)\le n/2-1. Delete the four diameter endpoints to obtain CC, of size m=n−4m=n-4. Every remaining point has a supporting radius-rr disk containing CC, centered at one of its rr-neighbors. Thus CC is strictly convex, and

m/2≤q(C)≤q(A)−1≤m/2.m/2\le q(C)\le q(A)-1\le m/2.

For n≥12n\ge12, the checked Fishburn equality classification makes CC a regular even polygon or a regular odd polygon minus one vertex. The former admits only its center as a possible added point by the preceding lemma; the latter admits only its missing vertex by the checked palette lemma. Neither admits the four deleted points with the required palette. The smaller even orders are covered by small_orders.

The square endpoint case

Suppose the only diameter pairs are the diagonals of a unit square, and the second-largest distance is one. All remaining points lie in

K={(x,y):x2+y2+∣x∣+∣y∣≤1/2}.K=\{(x,y):x^2+y^2+|x|+|y|\le1/2\}.

Set k=(3−1)/2k=(\sqrt3-1)/2 and s=2k=3−1s=2k=\sqrt3-1. The set KK lies in the radius-kk disk, meeting its boundary only at the four junctions (±k,0),(0,±k)(\pm k,0),(0,\pm k). Let hh of these junctions be present and let zz other points lie in the interior of KK. Since diam⁡K<1\operatorname{diam}K<1, each nonjunction boundary point has one unit corner neighbor, each junction has two, and interior points have none. Therefore

mA(1)=n+h−z.m_A(1)=n+h-z.

If this class is rich, z≤h−1z\le h-1. Under that richness assumption, if two opposite junctions are present, their distance ss is a second rare distance. There are at most two ss-pairs within KK, all opposite junction pairs. Each nonjunction boundary point has at most one corner neighbor at distance ss, each junction has none, and each interior point has at most two. Hence

mA(s)≤2+(n−4−z−h)+2z=n−2−h+z≤n−3.m_A(s)\le2+(n-4-z-h)+2z=n-2-h+z\le n-3.

To check those corner bounds, equal distances to two square corners put a point on an axis or a diagonal. At its intersections with ∂K\partial K, the repeated corner radii are 11, 2−3\sqrt{2-\sqrt3}, or 2−2\sqrt{2-\sqrt2}, none equal to ss. Three equidistant corners force the square center, with radius 1/2≠s1/\sqrt2\ne s.

The endpoint-specific bound shows that two disjoint diameter pairs and mA(r)=n+4m_A(r)=n+4 force their endpoints to form this square. Then h−z=4h-z=4 forces all four junctions, so the sharp n+4n+4 case is excluded for every order. For even nn, the same bound also excludes mA(r)=n+3m_A(r)=n+3. If the minimum cross-pair unit degree is one, equality in the endpoint bound gives full activity, which was excluded above. If that minimum is two, the endpoints form a square; now h−z=3h-z=3 gives h≥3h\ge3, hence two opposite junctions and the second rare distance. The bound excludes the remaining minimum-degree-zero case directly. Likewise, mA(r)=n+2m_A(r)=n+2 is excluded at even order when some endpoint has no unit neighbor among the other diameter pair: equality in its bound again gives full activity.

Remaining cases and checks

Few diameter edges with inactive points remain untreated. At odd order, even full second-distance activity leaves the convex core one distance above its classified minimum. No general conclusion for these cases is asserted.

The author checked the cover count, both arc arguments, the enclosing-disk test, the even-polygon averaging, the square equations and corner radii, and the equality applications; the argument is author-recorded. The classification inputs were checked in the supplied full Fishburn paper by the earlier referenced proofs.