Skip to main content
Log in

The k-metric dimension

  • Published:
Journal of Combinatorial Optimization Aims and scope Submit manuscript

Abstract

For an undirected graph \(G=(V,E)\), a vertex \(\tau \in V\) separates vertices u and v (where \(u,v\in V\), \(u\ne v\)) if their distances to \(\tau \) are not equal. Given an integer parameter \(k \ge 1\), a set of vertices \(L\subseteq V\) is a feasible solution, if for every pair of distinct vertices, uv, there are at least k distinct vertices \(\tau _{1},\tau _{2},\ldots ,\tau _{k}\in L\), each separating u and v. Such a feasible solution is called a landmark set, and the k-metric dimension of a graph is the minimal cardinality of a landmark set for the parameter k. The case \(k=1\) is a classic problem, where in its weighted version, each vertex v has a non-negative cost, and the goal is to find a landmark set with minimal total cost. We generalize the problem for \(k \ge 2\), introducing two models, and we seek for solutions to both the weighted version and the unweighted version of this more general problem. In the model of all-pairs (AP), k separations are needed for every pair of distinct vertices of V, while in the non-landmarks model (NL), such separations are required only for pairs of distinct vertices in \(V \setminus L\). We study the weighted and unweighted versions for both models (AP and NL), for path graphs, complete graphs, complete bipartite graphs, and complete wheel graphs, for all values of \(k \ge 2\). We present algorithms for these cases, thus demonstrating the difference between the two new models, and the differences between the cases \(k=1\) and \(k \ge 2\).

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Fig. 1
Fig. 2

Similar content being viewed by others

References

  • Adar R, Epstein L (2015) The weighted \(2\)-metric dimension of trees in the non-landmarks model. Discret Optim 17:123–135

    Article  MathSciNet  Google Scholar 

  • Babai L (1981) On the order of uniprimitive permutation groups. Ann Math 113(3):553–568

    Article  MathSciNet  MATH  Google Scholar 

  • Beerliova Z, Eberhard F, Erlebach T, Hall A, Hoffmann M, Mihalák M, Ram LS (2006) Network discovery and verification. IEEE J Sel Areas Commun 24(12):2168–2181

    Article  MATH  Google Scholar 

  • Blum M, Floyd RW, Pratt VR, Rivest RL, Tarjan RE (1973) Time bounds for selection. J Comput Syst Sci 7(4):448–461

    Article  MathSciNet  MATH  Google Scholar 

  • Cáceres J, Hernando MC, Mora M, Pelayo IM, Puertas ML, Seara C, Wood DR (2007) On the metric dimension of cartesian products of graphs. SIAM J Discret Math 21(2):423–441

    Article  MathSciNet  MATH  Google Scholar 

  • Chartrand G, Zhang P (2003) The theory and applications of resolvability in graphs: A survey. Congr Numer 160:47–68

    MathSciNet  MATH  Google Scholar 

  • Chartrand G, Eroh L, Johnson MA, Oellermann OR (2000) Resolvability in graphs and the metric dimension of a graph. Discret Appl Math 105(1–3):99–113

    Article  MathSciNet  MATH  Google Scholar 

  • Chen X, Wang C (2014) Approximability of the minimum weighted doubly resolving set problem. In: Cai Z, Zelikovsky A, Bourgeois AG (eds) COCOON. Lecture Notes in Computer Science, vol 8591. Springer, Berlin, pp 357–368

  • Chvátal V (1983) Mastermind. Combinatorica 3(3):325–329

    Article  MathSciNet  MATH  Google Scholar 

  • Díaz J, Pottonen O, Serna MJ, van Leeuwen EJ (2012) On the complexity of metric dimension. In: Epstein L, Ferragina P (eds) ESA. Lecture Notes in Computer Science, vol 7501. Springer, Berlin. pp 419–430

  • Epstein L, Levin A, Woeginger GJ (2015) The (weighted) metric dimension of graphs: Hard and easy cases. Algorithmica 72(4):1130–1171

    Article  MathSciNet  MATH  Google Scholar 

  • Estrada-Moreno A, Yero IG, Rodríguez-Velázquez JA (2014) \(k\)-metric resolvability in graphs. Electron Notes Discret Math 46:121–128

    Article  MathSciNet  MATH  Google Scholar 

  • Estrada-Moreno A, Rodríguez-Velázquez JA, Yero IG (2014) The \(k\)-metric dimension of a graph. CoRR. arXiv:1312.6840

  • Estrada-Moreno A, Yero IG, Rodríguez-Velázquez JA (2014) The \(k\)-metric dimension of corona product graphs. CoRR. arXiv:1401.3780

  • Estrada-Moreno A, Yero IG, Rodríguez-Velázquez JA (2014) The \(k\)-metric dimension of the lexicographic product of graphs. arXiv:1410.7287

  • Harary F, Melter R (1976) The metric dimension of a graph. Ars Comb 2:191–195

    MATH  Google Scholar 

  • Hartung S, Nichterlein A (2013) On the parameterized and approximation hardness of metric dimension. In: Umans C (ed), Proceedings of IEEE conference on computational complexity (CCC), pp 266–276

  • Hauptmann M, Schmied R, Viehmann C (2012) Approximation complexity of metric dimension problem. J Discret Algorithms 14:214–222

    Article  MathSciNet  MATH  Google Scholar 

  • Johnson D (1985) The NP-completeness column: an ongoing guide. J Algorithms 6(3):434–451

    Article  MathSciNet  MATH  Google Scholar 

  • Khuller S, Raghavachari B, Rosenfeld A (1996) Landmarks in graphs. Discret Appl Math 70(3):217–229

    Article  MathSciNet  MATH  Google Scholar 

  • Melter RA, Tomescu I (1984) Metric bases in digital geometry. Comput Vis Graph Image Process 25:113–121

    Article  MATH  Google Scholar 

  • Sebö A, Tannier E (2004) On metric generators of graphs. Math Oper Res 29(2):383–393

    Article  MathSciNet  MATH  Google Scholar 

  • Shanmukha B, Sooryanarayana B, Harinath KS (2002) Metric dimension of wheels. Far East J Appl Math 8(3):217–229

    MathSciNet  MATH  Google Scholar 

  • Slater PJ (1975) Leaves of trees. Congr Numer 14:549–559

    MathSciNet  MATH  Google Scholar 

  • Yero IG, Estrada-Moreno A, Rodríguez-Velázquez JA (2014) The \(k\)-metric dimension of a graph: complexity and algorithms. Corr. arXiv:1401.0342

Download references

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Leah Epstein.

Appendix

Appendix

1.1 The case \(n=5\) for complete wheels in AP and NL

Theorem 39

Consider the case \(n=5\).

  • For AP, \(md_{2}^{AP}(W_5)=4\), \(wmd_{2}^{AP}(W_5)=w(C)\), and \(md_{3}^{AP}(W_5)=md_{4}^{AP}(W_5)=\infty \).

  • For NL, \(md_{2}^{NL}(W_5)=3\), and

    $$\begin{aligned} wmd_{2}^{NL}(W_5)=\min \{w(C),w(V)-\max _{1 \le i \le 4} (w(c_{i})+w(c_{i+1}))\} \ . \end{aligned}$$

    Additionally, \(md_{3}^{NL}(W_5)=4\), and

    $$\begin{aligned} wmd_{3}^{NL}(W_5)=\min \{w(C), W(V)-\max _{1 \le i \le 4} w(c_{i}) \} \ . \end{aligned}$$

Proof

First, consider AP and \(k=2\). A pair of cycle vertices \(c_{i}\) and \(c_{i+2}\) cannot be separated by any vertex except for \(c_{i}\) and \(c_{i+2}\), as they have distances of 1 to all other vertices. Thus, all the cycle vertices (which can be split into two such pairs) must belong to any landmark set \(L \in LS_{2}^{AP}(W_5)\). We claim that \(C \in LS_{2}^{AP}(W_5)\), and it is a unique minimal landmark set. By Property 1, and the above explanation, it is sufficient to show two separations between \(c_{i}\) (for \(1\le i\le 4\)) and h. One separation results from \(c_{i}\) being a member of the landmark set. Additionally, \(d(h,c_{i+2})=1\) while \(d(c_{i},c_{i+2})=2\), so \(c_{i+2}\) separates \(c_{i}\) and h. Note that V is a landmark set as well, but it is not minimal.

Next, consider NL and \(k=2\), and let \(L \in LS_{2}^{NL}(W_5)\). In this case, for \(i=1,2\), at least one of \(c_{i}\) and \(c_{i+2}\) must belong to L (otherwise they cannot be separated, as explained above). If there is at least one index j such that \(c_{j}\notin L\) and, in addition, \(h\notin L\), then only \(c_{j+2}\) (if it is in L) separates \(c_{j}\) and h (as their distances to both \(c_{j-1}\) and \(c_{j+1}\) is 1), and \(L \notin LS_{2}^{NL}(W_5)\). Thus, if \(h\notin L\), then \(L=C\), where C is a trivial landmark set for NL. On the other hand, a set of the form \(\{c_{i},c_{i+1},h\} \in LS_{2}^{NL}(W_5)\), as \(c_{i-1}\) and \(c_{i+2}\) are separated by both \(c_{i}\) and \(c_{i+1}\). Therefore, a minimal landmark set (with respect to set inclusion) consists of either of C, or h and two adjacent cycle vertices.

Finally, we deal with \(k = 3\). For AP, due to the above, there cannot be more than two separations between \(c_{i}\) and \(c_{i+2}\), so \(md_{3}^{AP}(W_5)=\infty \). Since \(md_{4}^{AP}(W_n) \ge md_{3}^{AP}(W_5)\), we get \(md_{4}^{AP}(W_5)=\infty \).

Consider the case of NL and \(k=3\). We already showed that any landmark set (for NL and \(k=2\)) with at most three vertices must contain h and two cycle vertices. For such a set, the other two cycle vertices will only have two separations. Thus, for NL and \(k=3\) only trivial solutions exist. \(\square \)

1.2 Minimal landmark sets for complete wheels with \(6 \le n \le 8\) in AP

1.2.1 \(k=2\)

We discuss minimal landmark sets (with respect to set inclusion) for AP, \(n=6,7,8\) and \(k=2\). By Corollary 19, for \(n=8\), such sets do not contain h, and the sets of cycle vertices that satisfy Condition 1 have at least four (cycle) vertices. Any subset of at least five cycle vertices is a landmark set (for AP and \(k=2\)) , but only those with two consecutive non-landmarks are minimal, with respect to set inclusion (sets of the form \(c_{j},c_{j+1},\ldots ,c_{j+4}\) for \(1\le j\le 7\)), since if there is no pair of consecutive non-landmarks, there must be three consecutive landmarks \(c_{i-1},c_{i},c_{i+1}\), and \(c_{i}\) is redundant (i.e. it is a landmark set without it). Any subset of four cycle vertices for which there is no pair of consecutive non-landmarks satisfies Condition 1, and it has the form \(c_{i},c_{i+1},c_{i+3},c_{i+5}\) for some \(1\le i \le 7\).

For \(n=6,7\), a set satisfying Condition 1 has at least three cycle vertices, and we already considered such sets. The next corollary specifies all minimal landmark sets for \(n=6,7\), where unlike the case \(n\ge 8\), there exist such sets that contain h.

Corollary 40

Let \(L\subseteq C\), such that \(|L|\ge 4\). Then, \(L \in LS_{2}^{AP}(W_n)\), for \(n=6,7\). Furthermore, assume that \(X \subseteq C\) satisfies Condition 1, and \(|X|=3\). Then, for \(n=6,7\), \(Y=X\cup \{h\} \in LS_{2}^{AP}(W_n)\), while \(X \notin LS_{2}^{AP}(W_n)\).

Proof

For \(n=6,7\), any set with four cycle vertices (particularly L) satisfies Condition 1; for \(n=6\), there is at most one cycle vertex not in L, and for \(n=7\), there are at most two cycle vertices not in L, and Condition 1 is satisfied no matter whether they are adjacent or not. By Proposition 17, there are two separations for any pair of the form \(h,c_{i}\). By Lemma 18, any pair of cycle vertices also has at least two separations. Hence, \(L \in LS_{2}^{AP}(W_n)\) for \(n=6,7\).

The set X has two separations for any pair of cycle vertices, by Lemma 18, and in Y, a pair \(h,c_{j}\) is separated by h and \(X\setminus \{c_{j-1},c_{j+1}\}\), which must contain at least one vertex, so \(Y \in LS_{2}^{AP}(W_n)\) for \(n=6,7\). We show that \(X \notin LS_{2}^{AP}(W_n)\), for \(n=6,7\).

For \(n=7\), the only form of a set X that satisfies Condition 1 is \(c_{i},c_{i+2},c_{i+4}\), since in the case of two consecutive vertices not in X, all remaining cycle vertices must be in X. For this set, h and \(c_{i+1}\) are only separated by \(c_{i+4}\).

For \(n=6\), X can be of one of two forms. If X consists of \(c_{i},c_{i+1},c_{i+3}\), then \(c_{i+2}\) and h are only separated by \(c_{i}\). If X consists of \(c_{i},c_{i+1},c_{i+2}\), then \(c_{i+1},h\) are only separated by \(c_{i+1}\). Thus, \(X \notin LS_{2}^{AP}(W_n)\) for \(n=6,7\). \(\square \)

1.2.2 \(k=3\)

A minimal landmark set (with respect to set inclusion) for AP, \(n=7,8\) and \(k=3\) consists of \(n-2\) cycle vertices, and any subset of \(n-2\) vertices is a landmark set. For \(n=6\), a set that satisfies Condition 2 has at least four cycle vertices. We now specify the structure of minimal landmark sets (with respect to set inclusion) for \(n=6\), and show these sets are exactly all subsets of five vertices.

Corollary 41

For \(n=6\), \(k=3\) and AP, a set of four cycle vertices is not a landmark set. Any subset of V with five vertices is a landmark set.

Proof

A set of four (consecutive) cycle vertices \(c_{i-2},c_{i-1},c_{i+1},c_{i+2}\) only makes two separations for the following three pairs: \(h,c_{i-2}\), \(h,c_{i}\), \(h,c_{i+2}\). Any set of five vertices gives another separation to each one of the pairs, as both h and \(c_{i}\) separate each one of the three pairs. \(\square \)

1.2.3 \(k \ge 4\)

Theorem 42

For AP, and \(n\ge 6\), \(md_{k}^{AP}(W_n)=\infty \) for \(k\ge 5\). Additionally, \(md_{4}^{AP}(W_n)=n-1\) if \(n\ge 7\) (in this case \(wmd_{4}^{AP}(W_n)=w(C)\)), and \(md_{4}^{AP}(W_n)=6\) for \(n=6\) (in this case \(wmd_{4}^{AP}(W_n)=w(V)\)).

Proof

Since a pair of the form \(c_{i},c_{i+1}\) can be separated by exactly four vertices, a landmark set for \(k=4\) must contain all cycle vertices, and there is no valid landmark set for \(k\ge 5\). Any cycle vertex except for \(c_{i+1}\) and \(c_{i-1}\) separates h and \(c_{i}\), thus C is a landmark set for \(k=4\) if \(n\ge 7\). For \(n=6\), there are only three separations by cycle vertices between h and \(c_{i}\), and therefore V is the only landmark set (as h adds a separation). \(\square \)

1.3 Minimal landmark sets for complete wheels and \(6 \le n \le 8\) in NL

1.3.1 \(k=2\)

The minimal landmark sets (with respect to set inclusion) for \(n=8\) and \(k=2\) are not the same as for AP, since the condition is slightly weaker. All subsets of four cycle vertices where there is no pair of consecutive non-landmarks (that were defined for AP) are still minimal landmark sets. Minimal landmark sets with a pair of consecutive non-landmarks are possible too, and have the form \(c_{i},c_{i+1},c_{i+3},c_{i+4}\) (for some \(1\le i\le 7\)).

For \(n=6,7\), we consider sets with at least three cycle vertices.

Corollary 43

Let \(L\subseteq C\), such that \(|L|\ge 4\). Then, \(L \in LS_{2}^{NL}(W_n)\), for \(n=6,7\). Furthermore, assume that \(X \subseteq C\) satisfies Condition 3, and \(|X|=3\). Then, \(Y=X\cup \{h\} \in LS_{2}^{NL}(W_n)\), for \(n=6,7\), while \(X \in LS_{2}^{NL}(W_n)\) only if \(n=6\), and X consists of three consecutive cycle vertices.

Proof

We showed in Corollary 40 that any subset of four vertices is a landmark set for AP, \(n=6,7\) and \(k=2\), hence (as stated in Sect. 1) it is also a landmark set for NL, \(n=6,7\) and \(k=2\).

For \(n=7\), the proof that any landmark set (for NL and \(k=2\)) has at least four vertices is done in the same way as for AP.

For \(n=6\), the proof that a triplet of cycle vertices that are not consecutive is not a landmark set (for NL and \(k=2\)) is done in the same way as for AP (for the proof of these two properties for AP see Corollary 40).

If X consists of \(c_{i},c_{i+1},c_{i+2}\), then \(c_{i-1},h\) are separated by \(c_{i+1}\) and \(c_{i+2}\), while \(c_{i+3},h\) are separated by \(c_{i}\) and \(c_{i+1}\), and \(c_{i-1},c_{i+3}\) are separated by \(c_{i}\) and \(c_{i+2}\). \(\square \)

We find that a minimal landmark set (with respect to set inclusion) for \(n=7\) is the same as for AP. For \(n=6\), a minimal landmark set consists of either three consecutive cycle vertices, or h together with three cycle vertices that are not consecutive.

1.3.2 \(k=3,4\)

Lemma 44

For NL, \(n = 6\) and \(k \in \{3,4\}\), only trivial landmark sets exist (\(md_{k}^{NL}(W_n)=5\) for \(k=3\) or \(k=4\)).

For NL, \(n=7\) and \(k=3\), \(L\subseteq V\) is a landmark set if and only if it satisfies Condition 4 and \(|L|\ge 5\) (thus \(md_{3}^{NL}(W_n)=5\)). For NL, \(n=7\) and \(k=4\), \(L\subseteq V\) is a landmark set if and only if it either satisfies Condition 4 and in addition \(h\in L\), or \(L=C\) (thus \(md_{4}^{NL}(W_n)=5\)).

For NL, \(n=8\) and \(k=3\), \(L\subseteq C\) is a landmark set if and only if it satisfies Condition 4. In addition, \(md_{3}^{NL}(W_n)=5\).

For NL, \(n=8\) and \(k=4\), \(L\subseteq V\) is a landmark set if and only if it either satisfies Condition 4 and in addition \(h\in L\), or it has at least six cycle vertices (thus \(md_{4}^{NL}(W_n)=6\)).

Proof

For \(n=6\), a set L that satisfies Condition 4 has at least four cycle vertices. If \(c_{i},h \notin L\), then as \(c_{i-1},c_{i+1}\) do not separate \(c_{i}\) and h, there are two separations between them. Thus, L must contain another vertex, hence \(|L|\ge 5\), and it is a trivial landmark set for NL.

For \(n=7\), a set L that satisfies Condition 4 has at least four cycle vertices. If it has exactly four cycle vertices, then it has the form \(c_{i-2},c_{i-1},c_{i+1},c_{i+2}\). If \(h \notin L\), then similarly to the case \(n=6\), there are two separations between \(c_{i}\) and h. On the other hand if \(h \in L\), this is a landmark set for \(k=3,4\). Assume now that \(h \notin L\), and thus L must contain at least five cycle vertices. If \(c_{i} \notin L\), then there are three separations between h and \(c_{i}\), so L is a landmark set for \(k=3\) but not for \(k=4\), and the only landmark set for \(k=4\) not containing h is C.

For \(n=8\) and \(k=3\), a set that satisfies Condition 4 has at least five cycle vertices (using Proposition 33), and there are three separations for any pair of the form \(h,c_{i}\). By Lemma 34, any pair of non-landmark cycle vertices also has three separations. The set \(L=\{c_{1},c_{2},c_{4},c_{5},c_{7}\}\) is a landmark set with five vertices, thus \(md_{3}^{NL}(W_n)=5\).

For \(n=8\) and \(k=4\), a set L that satisfies Condition 4 has at least five cycle vertices. If \(h \in L\), then it is a landmark set. Otherwise, by the condition, if \(c_{i} \notin L\), then \(c_{i-1},c_{i+1} \in L\), and to obtain four separations between \(c_{i}\) and h, L must contain \(C\setminus \{c_{i}\}\), which results in a landmark set of this form. \(\square \)

1.3.3 \(k\ge 5\)

Theorem 45

Let \(k\ge 5\). For NL and \(n\ge 7\), \(md_{k}^{NL}(W_n)=n-2\) if \(n\ge k+4\), and \(md_{k}^{NL}(W_n)=n-1\) if \(n\le k+3\).

Proof

Since a pair of cycle vertices \(c_{i},c_{j}\) that are not landmarks can be separated by at most four vertices, any landmark set must contain all cycle vertices except for at most one. Any set consisting of \(C\setminus \{c_{i}\}\) for some \(1\le i\le n-1\) is a landmark set (for NL and \(k \ge 5\)) if and only if h and \(c_{i}\) can be separated by k cycle vertices. The number of cycle vertices that can separate them is \(n-4\) (all cycle vertices except for \(c_{i-1},c_{i},c_{i+1}\)). Thus, if \(k\ge n-3\), only trivial landmark sets exist, and otherwise the minimal landmark sets consist of all cycle vertices expect for one. \(\square \)

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Adar, R., Epstein, L. The k-metric dimension. J Comb Optim 34, 1–30 (2017). https://doi.org/10.1007/s10878-016-0073-1

Download citation

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s10878-016-0073-1

Keywords

Navigation