A directed graph (or digraph) consists of a finite vertex set V and a set of ordered edges E ⊆ V × V , each edge (u, v) indicating a one-way connection from u (source) to v (target). A bidirected graph is a generalization of an undirected graph where each edge is assigned a direction at each of its endpoints independently, allowing more expressive edge orientation. A multidirected graph is a structure with vertices and edges, where edges may repeat, sources and targets are assigned, and multiplicities recorded. A molecular graph models a molecule with atoms as vertices and bonds as edges, representing its structural connectivity. In this paper, we examine definitions such as molecular bidirected graphs and multidirected graphs. These are concepts that extend molecular graphs by incorporating directional information.
This section provides an overview of the fundamental concepts and definitions essential for the discussions in this paper. Throughout this work, all graphs are assumed to be finite and without loops.
Graph theory studies vertices and edges, modeling pairwise relationships, enabling analysis of networks, con-nectivity, paths, cycles, and structures
Definition 1 (Directed Graph).
• V is a finite set of vertices, and • E ⊆ V × V is a set of ordered pairs called directed edges (or arcs).
Each edge (u, v) ∈ E represents a connection directed from vertex u (the tail) to vertex v (the head).
Example 1 (Directed Graph example: Hydrogen chloride polarity (H→Cl)). (cf.
Thus G = (V,E) is a directed graph (digraph) whose single arc (H,Cl) encodes the orientation of the polar covalent bond from hydrogen (tail) to chlorine (head).
Definition 2 (Bidirected Graph).
• G is a simple undirected graph on vertex set V and edge set E.
• For each vertex-edge pair (v,e), +1, if e is directed toward v, τ(v, e) = −1, if e is directed away from v, 0, if v is not incident to e.
We call G the underlying graph and τ the bidirection function.
Example 2 (Bidirected Graph example: Ammonia–borane dative bond NH3 → BH3). (cf.
Define the bidirection function τ : V × E → {−1, 0, 1} by τ(N,e) = −1 and τ(B,e) = +1, with τ(v,e) = 0 for any nonincident pair (none here). Then B = (G, τ) is a bidirected graph in which the edge is locally oriented “away from” N and “toward” B, encoding the donor (N) → acceptor (B) dative interaction.
A multidirected graph is a structure with vertices and edges, where edges may repeat, sources and targets are assigned, and multiplicities recorded
Definition 3 (Multidirected Graph).
Example 3 (Multidirected Graph example: Carbon monoxide triple bond as multiplicity). (cf.
Define the multiplicity map m : V × V → N0 by m(C, O) = {e ∈ E : s(e) = C, t(e) = O} = 3, m(u, v) = 0 for all (u, v) ̸= (C, O).
Thus G encodes the C≡O triple bond as three parallel directed edges from C to O with consistent multiplicity.
A molecular graph models a molecule with atoms as vertices and bonds as edges, representing its structural connectivity
Definition 4 (Molecular Graph).
Example 4 (Molecular Graph example: Methane (CH4) connectivity). (cf.
Then G = (V,E) captures the tetrahedral connectivity of CH4 without imposing any edge orientation.
As the main outcome of this paper, we introduce definitions of new graph classes and examine their properties.
A molecular directed graph represents atoms as labeled vertices and chemical bonds as directed edges, capturing oriented molecular interactions.
Definition 5 (Molecular Directed Graph (MDG)). Fix finite attribute sets ΣV (vertex/atom attributes, e.g., element, charge, isotope) and ΣE (edge/bond attributes, e.g., bond order, stereo). A molecular directed graph is a tuple MD = (V,E,s, t, λV , λE),
• V is a finite set of vertices (atoms), • E is a finite set of edges (bonds), • s, t : E → V assign to each edge e ∈ E its source s(e) and target t(e), • λV : V → ΣV labels each vertex with its atomic attributes, • λE : E → ΣE labels each edge with its bond attributes.
We write the underlying directed edge set of MD as E→ := { (s(e), t(e)) | e ∈ E } ⊆ V × V.
Example 5 (Ammonia (NH3): three polar N–H bonds oriented Hi → N). (cf.
Label vertices and edges as λV (N) = N, λV (Hi) = H (i = 1, 2, 3), λE(ei) = single (i = 1, 2, 3).
The underlying directed edge set is E→ = {(H1, N), (H2, N), (H3, N)}, encoding the polarity of each N–H bond from hydrogen (source) to nitrogen (target).
Example 6 (Carbon dioxide (CO2): two polar C=O bonds oriented C → Oi). Let ΣV = {C, O} and ΣE = {double}. Define MD = (V,E,s, t, λV , λE)
V = {C, O1, O2}, E = {e1, e2}, s(e1) = C, t(e1) = O1, s(e2) = C, t(e2) = O2.
Vertex and edge labels are λV (C) = C, λV (Oi) = O (i = 1, 2), λE(e1) = λE(e2) = double.
Thus E→ = {(C, O1), (C, O2)}, representing both C=O bonds by directed edges from carbon to oxygen (one per oxygen). Multiplicity information (double bonds) is carried here only as an edge label.
Example 7 (Sodium chloride (NaCl): ionic interaction oriented Na → Cl). (cf.
Set the labels λV (Na) = Na, λV (Cl) = Cl, λE(e) = ionic.
Then E→ = {(N a, Cl)}, capturing the ionic character by a single directed edge pointing from sodium (donor/source) to chlorine (acceptor/target).
Example 8 (Water (polar O–H bonds as oriented edges)). (cf.
Thus E→ = {(H1, O), (H2, O)} captures the polarity (hydrogen → oxygen).
Example 9 (Hydrogen chloride (H→Cl bond polarity)). (cf.
V = {H,Cl}, E = {e}, s(e) = H, t(e) = Cl, λV (H) = H, λV (Cl) = Cl, λE(e) = single.
Then E→ = {(H,Cl)} represents the oriented H–Cl bond.
Theorem 1 (MDG generalizes a directed graph). Let Dig denote the class of finite directed graphs G = (V,E→) with E→ ⊆ V × V . There exist maps UD : MD 7→ (V,E→) ∈ Dig, ID : (V,E→) 7→ MD such that UD ◦ ID = idDig. Hence every directed graph is (canonically) a special case of a molecular directed graph.
Proof : Define the forgetful map UD(V, E, s, t, λV , λE) := (V, {(s(e), t(e)) : e ∈ E}) ∈ Dig.
Given any G = (V, E→) ∈ Dig, define the trivial-label embedding ID(G) := (V, E, s, t, λV , λE), by taking E := E→, s(u, v) := u, t(u, v) := v for each (u, v) ∈ E→, and choosing fixed elements v∗ ∈ ΣV , e∗ ∈ ΣE with λV (v) := v∗ for all v ∈ V, λE(e) := e∗ for all e ∈ E.
Then UD ID(G) = (V, {(s(u, v), t(u, v)) : (u, v) ∈ E→}) = (V, E→) = G, so UD ◦ ID = idDig. Therefore every directed graph is realized as an MDG with constant labels, and every MDG forgets to a directed graph via UD. □
A molecular bidirected graph models atoms as labeled vertices with edges independently directed at endpoints, encoding bidirectional or asymmetric bond orientations.
Definition 6 (Molecular BiDirected Graph (MBDG)). Fix finite attribute sets ΣV and ΣE. A molecular bidirected graph is a quintuple MB = (V,E,τ,λV , λE),
• V is a finite set of vertices and E a finite set of (undirected) edges, • τ : V × E → {−1, 0, 1} is a bidirection function satisfying:
τ(v,e) = 0 if v is not incident to e; τ(v, e) = +1 if e is directed toward v at v; τ(v, e) = −1 if e is directed away from v at v; • λV : V → ΣV and λE : E → ΣE assign molecular attributes to vertices/edges.
The underlying bidirected graph is (G, τ) with G = (V, E).
Example 10 (Ammonia–borane dative bond: NH3 → BH3 ). (cf.
Nonincident pairs map to 0 by definition. Endpoint signs encode “away from N” and “toward B”.
Example 11 (Diammine copper(II): two N→Cu donations). (cf.
τ(N1, e1) = −1, τ(Cu,e1) = +1, τ(N2, e2) = −1, τ(Cu,e2) = +1, (all other τ(·, ·) = 0), and labels λV (Cu) = Cu, λV (Ni) = N, λE(ei) = dative (i = 1, 2).
Each edge is oriented locally from N (donor) to Cu (acceptor) via τ.
Example 12 (Acetone–boron trifluoride adduct: O→B donation). (cf.
The bidirection function τ : V × E → {−1, 0, 1} is τ(Oac, e) = −1, τ(B,e) = +1, and τ(v,e′) = 0 for any nonincident pair (none here). Vertex/edge labels are λV (Oac) = O, λV (B) = B, λE(e) = dative.
At the oxygen endpoint the edge is locally “away” (−1), and at boron it is “toward” (+1), encoding the coordinate bond donation from the carbonyl oxygen of acetone to BF3.
Example 13 (Iron–carbonyl unit: σ donation and π backbonding (two bidirected edges)). (cf.
Define τ on incident pairs by τ(CO,eσ) = −1, τ(Fe, eσ) = +1 (CO → Fe σ-donation), τ(Fe, eπ) = −1, τ(CO,eπ) = +1 (Fe → CO π-backbonding), and set τ(·, ·) = 0 for all nonincident pairs. Labels are λV (Fe) = Fe, λV (CO) = CO, λE(eσ) = σ-donation, λE(eπ) = π-backbonding.
Thus the σ edge is locally oriented away from CO and toward Fe, while the π edge is locally oriented away from Fe and toward CO, capturing the classical synergic bonding picture of metal–carbonyl complexes within the bidirected framework.
Theorem 2 (MBDG generalizes a bidirected graph). Let BiDig denote the class of finite bidirected graphs B = (G, τ) with G = (V,E). There exist maps UB : MB 7→ (G, τ) ∈ BiDig, IB : (G, τ) 7→ MB such that UB ◦ IB = idBiDig.
Proof : Define the forgetful map UB(V, E, τ, λV , λE) := ((V, E), τ). Given B = ((V, E), τ) ∈ BiDig, fix v∗ ∈ ΣV , e∗ ∈ ΣE and define IB(B) := (V, E, τ, λV , λE), λV (v) := v∗ ∀v ∈ V, λE(e) := e∗ ∀e ∈ E.
Then UB(IB(B)) = ((V, E), τ) = B, hence UB ◦IB = idBiDig. Thus every bidirected graph appears as an MBDG with constant labels, and every MBDG forgets to a bidirected graph. □
A molecular multidirected graph extends directed models by allowing multiple labeled directed edges between atoms, capturing multiplicities and parallel chemical interactions.
Definition 7 (Molecular MultiDirected Graph (MMDG)). Fix finite attribute sets ΣV and ΣE. A molecular multidirected graph is a septuple MM = (V,E,s, t, m, λV , λE),
• V is a finite set of vertices, E a finite set of edges, • s, t : E → V assign the source and target of each edge, • m : V × V → N0 records multiplicities and is required to satisfy the consistency constraint m(u, v) = { e ∈ E : s(e) = u, t(e) = v } for all (u, v) ∈ V × V,
• λV : V → ΣV and λE : E → ΣE assign molecular attributes.
The underlying multidirected graph is (V,E,s, t, m).
Example 14 (Dimethyl sulfoxide S=O double bond as parallel directed edges). Let ΣV = {S, O}, ΣE = {σ, π} and define MM = (V,E,s, t, m, λV , λE) by V = {S, O}, E = {eσ, eπ}, s(eσ) = s(eπ) = S, t(eσ) = t(eπ) = O, λV (S) = S, λV (O) = O, λE(eσ) = σ, λE(eπ) = π.
Then the multiplicity function satisfies m(S, O) = {e ∈ E : s(e) = S, t(e) = O} = 2, m(u, v) = 0 otherwise.
Thus the S=O double bond is modeled by two parallel directed edges S → O.
Example 15 (Carbon monoxide C≡O triple bond as three parallel directed edges). Let ΣV = {C, O}, ΣE = {σ, π1, π2} and define MM = (V,E,s, t, m, λV , λE)
V = {C, O}, E = {eσ, eπ1 , eπ2 }, s(e) = C, t(e) = O for all e ∈ E, λV (C) = C, λV (O) = O, λE(eσ) = σ, λE(eπ1 ) = π1, λE(eπ2 ) = π2.
Hence m(C, O) = 3, m(u, v) = 0 otherwise, capturing a triple bond as three parallel directed edges in the same orientation C → O.
Example 16 (Carbon dioxide: two C=O double bonds as four parallel directed edges). (cf.
Assign sources and targets s(eσ,i) = C, t(eσ,i) = Oi, s(eπ,i) = C, t(eπ,i) = Oi (i = 1, 2), so each C–O bond is represented by two parallel directed edges C → Oi. Label vertices and edges by λV (C) = C, λV (Oi) = O (i = 1, 2), λE(eσ,i) = σ, λE(eπ,i) = π.
The multiplicity function m : V × V → N0 satisfies m(C, O1) = {e ∈ E : s(e) = C, t(e) = O1} = 2, m(C, O2) = 2, m(u, v) = 0 otherwise.
Thus the linear O=C=O structure is captured as two C=O double bonds, each modeled by a σ and a π edge in the same orientation C → Oi.
Example 17 (Metal–carbonyl synergy: CO→Ni σ donation and Ni→CO π backbonding). (cf.
Assign directions to encode the synergic interaction:
s(eσ) = CO, t(eσ) = Ni (σ donation CO → Ni), s(eπ) = N i, t(eπ) = CO (π backbonding Ni → CO). Set labels λV (Ni) = Ni, λV (CO) = CO, λE(eσ) = σ-donation, λE(eπ) = π-backbonding.
The multiplicity map is then m(CO,Ni) = {e ∈ E : s(e) = CO, t(e) = Ni} = 1, m(N i, CO) = {e ∈ E : s(e) = N i, t(e) = CO} = 1, m(u, v) = 0 for all other ordered pairs (u, v).
Hence this MMDG simultaneously represents two directed interactions between the same pair of vertices with opposite orientations and distinct edge labels, reflecting the classical σ/π synergy in metal–carbonyl bonding.
Theorem 3 (MMDG generalizes a multidirected graph). Let MultiDig denote the class of finite multidirected graphs G = (V,E,s, t, m) satisfying m(u, v) = |{e ∈ E : s(e) = u, t(e) = v}|. There exist maps UM : MM 7→ (V,E,s, t, m) ∈ MultiDig, IM : (V,E,s, t, m) 7→ MM such that UM ◦ IM = idMultiDig.
Proof : Define the forgetful map UM (V, E, s, t, m, λV , λE) := (V, E, s, t, m).
Given (V, E, s, t, m) ∈ MultiDig, choose fixed v∗ ∈ ΣV , e∗ ∈ ΣE and put IM (V, E, s, t, m) := (V, E, s, t, m, λV , λE), λV (v) := v∗ ∀v ∈ V, λE(e) := e∗ ∀e ∈ E.
Then UM (IM (V, E, s, t, m)) = (V, E, s, t, m), so UM ◦ IM = idMultiDig. Hence every multidirected graph is a special case of an MMDG with constant labels, and every MMDG forgets to a multidirected graph. □
Lemma 1 (Conservative extension). In each of the three settings (MDG, MBDG, MMDG), taking ΣV and ΣE to be singletons yields exactly the corresponding classical structure (directed, bidirected, multidirected) on the same carrier sets.
Proof : Let ΣV = {∗} and ΣE = {⋄}. Then all label maps λV , λE are uniquely determined and carry no additional information. The defining tuples reduce, under UD, UB, UM , to the classical data (V, E→), ((V, E), τ), and (V, E, s, t, m) respectively, with no loss or addition beyond the classical structures. Thus the molecular variants are conservative extensions. □
Lemma 2 (Exact retrieval via forgetful maps). For each molecular structure M ∈ {MD, MB, MM }, the corresponding forgetful map U ∈ {UD, UB, UM } returns the exact classical object obtained by discarding labels and keeping only the incidence/orientation data.
Proof : Immediate from the definitions of UD, UB, UM , which simply erase λV , λE and retain the structural maps (either (s, t), or τ, together with E and V ). The equalities UD(MD) = (V, E→), UB(MB) = ((V, E), τ), UM (MM ) = (V, E, s, t, m) hold by construction. □
In this paper, we examined definitions such as molecular bidirected graphs and multidirected graphs. In future work, we aim to conduct quantitative analyses of the proposed concepts through computational experiments. We also plan to explore possible extensions employing Fuzzy Graphs
No external funding or financial support was provided for this study.
The authors wish to thank all colleagues and mentors whose feedback and encouragement enriched this work. We are grateful to the community of researchers whose foundational contributions informed our developments. Special appreciation goes to the institutions that offered resources and technical infrastructure throughout this project.
This manuscript presents purely conceptual work without empirical data. Scholars interested in these ideas are invited to undertake experimental or case-study research to substantiate and extend the proposed frameworks.
This paper involves no human or animal subjects and thus did not require ethics committee review or approval.
We use generative AI and AI-assisted tools for tasks such as English grammar checking, and We do not employ them in any way that violates ethical standards.
The authors declare that there are no competing interests concerning the content or publication of this article.
The theoretical models and propositions herein have not yet been subjected to practical validation. Readers should independently verify all citations and be aware that inadvertent inaccuracies may remain. The opinions expressed are those of the authors and do not necessarily represent the views of affiliated organizations.