Asymmetric Tropical Distance

• Asymmetric Tropical Distance is a non-symmetric metric defined on the tropical torus that encodes orientation-dependent costs and underlies tropical convexity and optimization.
• It satisfies key metric axioms such as nonnegativity and a modified triangle inequality, and its structure enables efficient O(n) computation and polyhedral Voronoi constructions.
• The distance is crucial in applications including consensus methods, facility location, and phylogenetic analysis, bridging tropical geometry with directed metric theory.

The asymmetric tropical distance is a canonical non-symmetric distance function on the tropical torus $\mathbb{R}^n/\mathbb{R}\mathbf{1}$, playing a central role in tropical geometry, tropical convexity, the modeling of directed metrics, and applications such as phylogenetic tree space and location theory. Unlike its symmetric counterpart, the asymmetric tropical distance encodes orientation-dependent “costs” intrinsic to the combinatorial structure of tropical and polyhedral spaces, and is deeply connected to optimization, consensus, and clustering problems in non-Euclidean settings.

1. Definition and Basic Formulation

The tropical torus is the quotient $$\mathbb{R}^n/\mathbb{R}\mathbf{1} = \{ x + \mathbb{R}1 : x\in\mathbb{R}^n\}$$, where $1=(1,\dots,1)$. The asymmetric tropical distance $d_T(x,y)$ between $x, y\in\mathbb{R}^n$ is defined by

$$d_T(x,y) = (x, y) = \sum_{i=1}^n (y_i - x_i) + n \max_{1\le i \le n} (x_i - y_i).$$

This is invariant under simultaneous addition of constants to both vectors: for any $\lambda, \mu\in\mathbb{R}$,

$d_T(x+\lambda 1, y+\mu 1) = d_T(x, y),$

so it induces a well-defined quasi-metric on the tropical torus $\mathbb{R}^n/\mathbb{R}1$ (Lenzen et al., 27 Feb 2026). Equivalently, letting $\Delta = \operatorname{conv}\{e_1, \dots, e_n\}$ denote the standard $$\mathbb{R}^n/\mathbb{R}\mathbf{1} = \{ x + \mathbb{R}1 : x\in\mathbb{R}^n\}$$0-simplex, the distance can be described as the gauge:

$$\mathbb{R}^n/\mathbb{R}\mathbf{1} = \{ x + \mathbb{R}1 : x\in\mathbb{R}^n\}$$1

Alternative formulas and equivalent definitions reflecting its geometric origins appear in the literature (Comăneci et al., 2022, Lenzen et al., 27 Feb 2026):

$$\mathbb{R}^n/\mathbb{R}\mathbf{1} = \{ x + \mathbb{R}1 : x\in\mathbb{R}^n\}$$2

2. Structural Properties and Metric Axioms

The asymmetric tropical distance $$\mathbb{R}^n/\mathbb{R}\mathbf{1} = \{ x + \mathbb{R}1 : x\in\mathbb{R}^n\}$$3 exhibits the following properties:

• Nonnegativity and Positive Definiteness: $$\mathbb{R}^n/\mathbb{R}\mathbf{1} = \{ x + \mathbb{R}1 : x\in\mathbb{R}^n\}$$4, with equality if and only if $$\mathbb{R}^n/\mathbb{R}\mathbf{1} = \{ x + \mathbb{R}1 : x\in\mathbb{R}^n\}$$5 and $$\mathbb{R}^n/\mathbb{R}\mathbf{1} = \{ x + \mathbb{R}1 : x\in\mathbb{R}^n\}$$6 differ by a scalar, i.e., $$\mathbb{R}^n/\mathbb{R}\mathbf{1} = \{ x + \mathbb{R}1 : x\in\mathbb{R}^n\}$$7 (Lenzen et al., 27 Feb 2026).
• Triangle Inequality: $$\mathbb{R}^n/\mathbb{R}\mathbf{1} = \{ x + \mathbb{R}1 : x\in\mathbb{R}^n\}$$8 satisfies

$$\mathbb{R}^n/\mathbb{R}\mathbf{1} = \{ x + \mathbb{R}1 : x\in\mathbb{R}^n\}$$9

for all $1=(1,\dots,1)$0. Thus, it is a convex gauge and, modulo equivalence classes, defines a quasi-metric (Comăneci, 2023).

• Asymmetry: In general, $1=(1,\dots,1)$1. The maximal ratio $1=(1,\dots,1)$2 controls the skewness (Lenzen et al., 27 Feb 2026).
• Pseudo–Triangle Inequality: For all $1=(1,\dots,1)$3,

$1=(1,\dots,1)$4

• Relation to Symmetric Tropical Distance: If $1=(1,\dots,1)$5, then $1=(1,\dots,1)$6 (Lenzen et al., 27 Feb 2026, Comăneci et al., 2022).

The lack of symmetry is not merely a technical artifact; it encodes directed “effort” or “cost” and is crucial for consensus and optimization applications.

3. Geometric and Polyhedral Aspects

The balls of the asymmetric tropical distance centered at $1=(1,\dots,1)$7 are translates of $1=(1,\dots,1)$8: the unit ball is a product of a standard simplex and the lineality space. Consequently, $1=(1,\dots,1)$9 defines polyhedral, non-centrally symmetric geometry on the tropical torus (Comăneci et al., 2022). Notably:

• Voronoi Cells and Power Diagrams: Voronoi regions for $d_T(x,y)$0 are intersections of tropical halfspaces and are always contractible, manifesting as tropical polyhedra in the “max-tropical” sense. When “super-discreteness” holds (i.e., projections to coordinate axes are discrete), all cells remain globally polyhedral (Comăneci et al., 2022).
• Tropicalization of Classical Structures: The asymmetric tropical Voronoi diagram of a “super-discrete” site set arises as the tropicalization of an ordinary power diagram over real Puiseux series, preserving poset of intersections and combinatorial type (Comăneci et al., 2022).
• Delone Complexes: The clique complex of the dual graph of these Voronoi diagrams supports applications in the minimal free resolution of Laurent monomial modules.

4. Directed Metrics, Semimetrics, and Tropical Algebra

Any finite semimetric (asymmetric metric) $d_T(x,y)$1 on a set $d_T(x,y)$2 can be realized via the residuation operator in tropical algebra. Label $d_T(x,y)$3, and define the tropical matrix $d_T(x,y)$4. Then $d_T(x,y)$5 is tropically idempotent if $d_T(x,y)$6, and this idempotency is equivalent to the triangle inequality for $d_T(x,y)$7 (Johnson et al., 2012). The residuation distance,

$d_T(x,y)$8

represents an “asymmetric Hilbert metric” and is a canonical example of an asymmetric tropical distance (Johnson et al., 2012). The connection to tight spans shows that every finite directed semimetric can be embedded in a tropical polytope generated by $d_T(x,y)$9 points in tropical $x, y\in\mathbb{R}^n$0-space (Hirai et al., 2010, Johnson et al., 2012):

• Tight Span Construction: For a finite set $x, y\in\mathbb{R}^n$1, and a directed metric $x, y\in\mathbb{R}^n$2, the tight span $x, y\in\mathbb{R}^n$3 can be endowed with the asymmetric tropical $x, y\in\mathbb{R}^n$4-distance. The formula $x, y\in\mathbb{R}^n$5 links tropical convex geometry and classical polyhedral theory (Hirai et al., 2010).

5. Tropical Convexity, Location Theory, and Optimization

The asymmetric tropical distance is a gauge generating geodesic, star-convex, and polytropic structures in the tropical torus. Key analytic consequences include:

• Tropical $x, y\in\mathbb{R}^n$6 Gauges: Taking $x, y\in\mathbb{R}^n$7 for $x, y\in\mathbb{R}^n$8 yields convex functions whose $x, y\in\mathbb{R}^n$9 case recovers $$d_T(x,y) = (x, y) = \sum_{i=1}^n (y_i - x_i) + n \max_{1\le i \le n} (x_i - y_i).$$0 (Comăneci, 2023).
• Consensus, Medians, and Fermat-Weber Points: The tropical Fermat–Weber and median problems with respect to $$d_T(x,y) = (x, y) = \sum_{i=1}^n (y_i - x_i) + n \max_{1\le i \le n} (x_i - y_i).$$1 and its variants admit polyhedral solution sets; the minimizer is always in the tropical convex hull of the data (Comăneci et al., 2022, Comăneci, 2023, Comăneci et al., 2022). The computation reduces to a linear or transportation problem, and the solution set forms a polytrope, i.e., a polytope respecting both ordinary and tropical convexity (Comăneci et al., 2022, Comăneci et al., 2022).
• Phylogenetic Applications: In the Bergman fan $$d_T(x,y) = (x, y) = \sum_{i=1}^n (y_i - x_i) + n \max_{1\le i \le n} (x_i - y_i).$$2 (the tropical moduli of $$d_T(x,y) = (x, y) = \sum_{i=1}^n (y_i - x_i) + n \max_{1\le i \le n} (x_i - y_i).$$3-leaf equidistant trees), $$d_T(x,y) = (x, y) = \sum_{i=1}^n (y_i - x_i) + n \max_{1\le i \le n} (x_i - y_i).$$4 distinguishes distinct tree shapes, quantifies the cost of moving between tree topologies, and enables consensus/supertree constructions via tropical medians (Lenzen et al., 27 Feb 2026, Comăneci et al., 2022).

6. Computational Complexity and Algorithms

The explicit formula of $$d_T(x,y) = (x, y) = \sum_{i=1}^n (y_i - x_i) + n \max_{1\le i \le n} (x_i - y_i).$$5 allows $$d_T(x,y) = (x, y) = \sum_{i=1}^n (y_i - x_i) + n \max_{1\le i \le n} (x_i - y_i).$$6 time computation per pair (Lenzen et al., 27 Feb 2026). For key optimization problems:

• Transportation Problems: Fermat-Weber medians under $$d_T(x,y) = (x, y) = \sum_{i=1}^n (y_i - x_i) + n \max_{1\le i \le n} (x_i - y_i).$$7 can be computed as optimal points of a transportation linear program. The problem is always feasible with polyhedral solution space (Comăneci et al., 2022).
• Location Problems: In tropical convex hull optimization, the tropical center, splitter, and center are computable via (tropical) linear optimization or LP reduction (Comăneci, 2023).
• Voronoi and Power Diagrams: By tropicalization of classical power diagrams, one obtains efficient algorithms for constructing Voronoi cells in expected $$d_T(x,y) = (x, y) = \sum_{i=1}^n (y_i - x_i) + n \max_{1\le i \le n} (x_i - y_i).$$8 time for $$d_T(x,y) = (x, y) = \sum_{i=1}^n (y_i - x_i) + n \max_{1\le i \le n} (x_i - y_i).$$9 sites in $\lambda, \mu\in\mathbb{R}$0, markedly better than for the symmetric tropical distance (Comăneci et al., 2022).

7. Applications and Consensus in Phylogenetics

The asymmetric tropical distance underlies robust procedures for consensus and location in tree-space and tropical geometry:

• Phylogenetic Consensus and Supertrees: Consensus and supertree methods based on minimizing total $\lambda, \mu\in\mathbb{R}$1-distance inherit essential combinatorial properties—Pareto, co-Pareto, and super-majority clade rules—and remain within the tropical convex hull of the data (Comăneci, 2023, Comăneci et al., 2022, Comăneci et al., 2022).
• Clustering and Clade Structure: $\lambda, \mu\in\mathbb{R}$2 separates clusters by tree shape, is sensitive to directed changes, and supports clustering algorithms for collections of ultrametric trees (Lenzen et al., 27 Feb 2026).
• Location Theoretic Properties: The optimal solution to tropical convex location problems with $\lambda, \mu\in\mathbb{R}$3 is always tropically convex and, in many cases, strictly so. This provides theoretical justification for tropical methods in facility location, median identification, and consensus under non-symmetric dissimilarity (Comăneci, 2023).

---

The asymmetric tropical distance thus provides a unifying non-symmetric gauge for tropical convexity, optimization, directed metric geometry, and applications in combinatorial and statistical phylogenetics, supporting algorithmic advances and theoretical insights into non-Euclidean, polyhedral, and combinatorial data structures (Lenzen et al., 27 Feb 2026, Comăneci, 2023, Comăneci et al., 2022, Hirai et al., 2010, Comăneci et al., 2022, Comăneci et al., 2022, Johnson et al., 2012).