Maximum Parsimony¶
Select the branching tree that minimizes the total character-state-change cost needed to explain observed leaf states under a stated scoring model.
Core Idea¶
Maximum parsimony (MP) is a character-based tree-selection criterion. Given the same observed objects at the leaves and comparable character states, each candidate branching tree receives the minimum total cost of changes required to assign states to its internal nodes. MP selects the tree or tied trees with the lowest resulting score under a declared change-cost model. The word “maximum” refers to maximizing parsimony—equivalently minimizing the implied change score—not to maximizing the number of mutations.[1]
Scoring an already chosen tree is the small-parsimony problem. Searching among tree topologies is the additional MP problem. Fitch's familiar equal-cost procedure is one way to solve the first part; weighted or asymmetric transitions require a different score calculation. Neither a particular tree-search heuristic nor the assumption that the lowest-score tree is historically true is constitutive of the criterion.[1][2]
Structural Signature¶
Sig role-phrases: observed leaf characters → candidate branching trees → declared change costs → minimum-score selection.
- Observed leaf characters. The same taxa or witnesses have coded states for comparable positions or traits. Without observed characters, no change score discriminates candidate trees.[1][3]
- Candidate branching trees. Alternative topologies relate those leaves. With one fixed topology alone, the task is small parsimony rather than MP tree selection.[1]
- Declared change-cost model. For each tree and character, internal states are assigned to minimize transition cost. Equal-cost substitutions are one case; weighted or asymmetric costs change the score and may make rooting material.[1]
- Minimum-score selection. Sum the optimized character costs, then retain the globally lowest-scoring candidate tree or tied minimizers. A heuristic's best visited tree is only an approximation unless global optimality is certified.[1]
An exhaustive search, a particular matrix representation, and a claim about the true genealogy are not extra necessary roles.
What It Is Not¶
MP is not the broad maxim “prefer a simpler explanation.” Live Parsimony (Occam's Razor) is methodological kin, but MP specifies a character-change objective and a tree choice set. It is not maximum likelihood, which compares model probabilities, nor neighbor joining, which uses pairwise distances. Those methods can produce the same topology by coincidence without instantiating MP.[1][3]
A low MP score also does not prove a true tree. Felsenstein identifies parameter regimes where parsimony methods can converge on an incorrect phylogeny as data accumulate. That conditional inconsistency is a limitation of inference, not a component of the MP rule.[2]
Scope of Application¶
In phylogenetics, leaves may be species or sequences and characters may be aligned nucleotide positions or morphological traits. A candidate tree is scored by reconstructing changes along its branches. Carmel and colleagues distinguish a fixed-tree score from the optimization over all candidate phylogenies, including cases with asymmetric substitution costs where rooting matters.[1]
In textual criticism, leaves can be manuscript witnesses and characters their variant readings. Hyytiäinen's Acts 5 study encoded 54 manuscripts across 279 variation places and applied MP alongside neighbor joining. The resulting unrooted tree is an inference under the coding and tree assumptions, not a rooted stemma or proof that every copying history is tree-like.[3]
Other comparative datasets may use the criterion if objects, characters, candidate trees and change costs can be stated. A reticulate history or highly convergent characters can make a single tree a poor causal account even though the MP score remains computable.
Clarity¶
The adjective parsimonious can refer either to a fixed tree's optimized score or to the best topology under that score. Keeping small and maximum parsimony separate prevents a program that merely labels internal nodes on one tree from being credited with choosing among trees. Likewise, “fewest changes” is incomplete until the character coding and transition costs are given. Fitch's simple count and a weighted Sankoff-style score can rank trees differently.[1]
Rooted and unrooted formulations are not interchangeable under every cost scheme. In the source's treatment, symmetric costs admit the usual unrooted modeling convention, while asymmetric transitions make direction and root placement meaningful.[1]
Manages Complexity¶
MP compresses many possible ancestral assignments into one optimized score per candidate topology, then compares those scores under a common rule. This allows a large character matrix to inform a single tree-choice criterion while preserving the distinction between score calculation and topology search. The compression is computational, not epistemic: if characters were misaligned or a tree model is inappropriate, a concise score can hide those problems rather than solve them.[1][3]
Abstract Reasoning¶
Given a character matrix, one can ask which alternative tree requires the least change under the same declared cost model. That permits a repeatable comparison: alter the coding or cost matrix, recompute the internal-state minima, and inspect whether the preferred topology changes. A change in result then points to model sensitivity instead of an unexplained disagreement about “simplicity.”[1]
If a search program reports one low-score tree, the further inference “this is the MP optimum” requires a search guarantee. NP-hardness of the general tree-search problem makes heuristic exploration common, but the mathematical criterion and the algorithm's coverage are separate claims.[1]
Knowledge Transfer¶
Biological sequences and manuscript variants are unlike carriers, yet their observed characters, candidate trees, transition costs and minimum-score rule map onto the same operation. That is literal transfer of MP within comparative tree reconstruction, not an analogy based only on preferring simple narratives. Beyond such character-coded histories, live Optimization carries the general choice-set/objective skeleton; the named MP criterion does not automatically apply to every optimization task.[1][3]
Examples¶
Molecular phylogeny. The observed leaf characters are aligned sequence positions from the taxa; candidate branching trees place those taxa at leaves; a change-cost model assigns equal or weighted substitution costs and minimizes internal states on each tree; minimum-score selection retains the least-cost topology or a tie. Fitch's Hamming case is a qualified scoring procedure, not a universal necessity.[1]
Mapped back: the four roles are filled. Whether a heuristic found the global minimum is a separate search-audit question.
Acts 5 manuscript tree. The observed characters are 279 coded variation places across 54 manuscript witnesses; candidate trees represent alternative branching textual histories; the study's encoded readings supply a change-scoring basis; MP applies the minimum-score tree rule and compares its result with another method. No biological inheritance is required, though contamination among manuscripts can complicate tree interpretation.[3]
Mapped back: different artifacts fill the leaf and character roles, while the tree-score/argmin structure stays intact.
Negative boundary. A researcher computes the least ancestral change on one prespecified tree but never compares alternative topologies. That is small parsimony, not the full MP selection criterion.[1]
Structural Tensions¶
- Simple change history versus plausible repeated change. Minimizing change yields a clear decision rule, but parallel changes or reversals can be real. Leaning entirely on score can favor a misleading history; replacing it without a stated alternative makes comparison opaque. Diagnostic: Are unequal rates or repeated states plausible enough that the lowest-score tree should be checked against another inference model?[2]
- Global optimum versus feasible search. Exhaustive topology comparison can certify the MP minimum but grows hard; heuristics can reach useful trees without proving none scores better. Diagnostic: Is the reported tree a proven minimizer, or only the best found under a stated search effort?[1]
- One shared coding versus setting-specific evidence. Comparable characters make scoring reproducible, yet aggressive recoding can erase meaningful distinctions among molecular sites or manuscript readings. Diagnostic: Which scoring and missing-data choices materially alter the selected topology?[1][3]
Structural–Framed Character¶
Maximum Parsimony is mixed-structural, leaning formal: the argmin rule is exact once candidate trees, characters and costs are fixed, while those inputs are chosen for a historical inference task. Its evaluative weight is limited; “most parsimonious” means lowest score under the model, not demonstrably true or scientifically best. It is partly human-practice-bound as an inferential procedure over constructed datasets, though the resulting score is mathematically determined. Its institutional origin is comparative phylogenetics and related stemmatic methods, not a convention that makes a history simple. Its vocabulary travel reaches unlike character-coded tree problems; it does not become a generic label for every simple account. Import versus recognition requires the explicit tree, character and change-cost roles, not an aesthetic preference for fewer assumptions.
Live Optimization supplies the portable choice-set/objective skeleton: compare feasible alternatives under a declared score. MP adds ancestral-state reconstruction and minimum-change tree scoring. Its character: a formal comparative-inference criterion whose exact operation transfers across biological and textual carriers, while its distinctive semantics remain tied to character-coded branching histories.
Structural Core vs. Domain Accent¶
What is skeletal. A set of candidate solutions is evaluated by an objective and the minimum is selected. Live Optimization carries that cross-domain relation and is the proposed strict parent. It does not supply the particular way a candidate tree obtains its score.
What is domain-bound. The feasible choices are branching trees over common observed leaves; characters have states at those leaves; internal states are optimized under a specified transition cost; and total score ranks the trees. Remove either characters or competing trees and the MP criterion becomes a different task. Molecular versus manuscript carriers, Fitch versus weighted scoring, and exhaustive versus heuristic search vary without changing that core; historical truth and statistically consistent recovery are additional claims, not ingredients of the score.
Why this is not a prime. Optimization is recognized across many substrates. MP is recognized where the tree-and-character reconstruction test holds, even if the objects are DNA sequences in one setting and textual witnesses in another. Applying “maximum parsimony” to an ordinary budget decision imports the phrase but not its change-count tree objective. The parent carries general optimization reach; the named criterion keeps its comparative-history commitments.
Instantiates / Related Primes¶
This entry is a kind of Optimization.
DAG parent: live Optimization (Optimization). The tree set, change-score objective and minimum rule match its defined choice/objective/constraint structure. Live Parsimony (Occam's Razor) is related but weaker as a taxonomic candidate: its current broad simplicity preference does not supply the exact character-change objective. Live Optimality Criterion is a statistical-model scoring neighbor whose present definition is not the nearest tree criterion genus.
Relationships to Other Abstractions¶
Current abstraction Maximum Parsimony Domain-specific
Parents (1) — more general patterns this builds on
-
Maximum Parsimony is a kind of Optimization Prime
Maximum parsimony minimizes an explicit character-change objective over candidate trees.Live Optimization requires a feasible choice set, objective, constraints, and optimality rule. Here trees over fixed observed leaves are candidates, optimized ancestral-state change cost is the objective, character and topology assumptions constrain admissibility, and the selected topology is a global minimum or an explicitly approximate search result.
Hierarchy path (1) — routes to 1 parentless root
- Maximum Parsimony → Optimization
Neighborhood in Abstraction Space¶
Maximum Parsimony sits in a sparse region of the domain-specific corpus (79th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Codes, Matrices & Combinatorial Problems (30 abstractions)
Nearest neighbors
- Optimality criterion — 0.84
- Branch Decomposition — 0.83
- Suffix Tree — 0.82
- De Novo Transcriptome Assembly — 0.82
- Phylogenetic nomenclature — 0.82
Computed from structural-signature embeddings · 2026-10-08
Not to Be Confused With¶
- Small parsimony. It optimizes internal character assignments on a fixed tree. Tell: is the topology also chosen among competitors?[1]
- Maximum likelihood. It ranks trees by a probability model rather than minimum implied character-change cost. Tell: is the objective a transition-count/cost or a likelihood?[1]
- Simple narrative preference. It may invoke Occam's Razor without character coding. Tell: can two candidate trees be scored under one declared change model?
References¶
[1] Amir Carmel, Noa Musa-Lempel, Dekel Tsur and Michal Ziv-Ukelson, "The Worst Case Complexity of Maximum Parsimony", Journal of Computational Biology 21(11), 799–808 (2014), §1.1 Problems 1–2. Directly supports the fixed-tree/tree-search distinction, score variants and computational limits. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s ↩t
[2] Joseph Felsenstein, "Cases in Which Parsimony or Compatibility Methods Will Be Positively Misleading", Systematic Zoology 27(4), 401–410 (1978), publisher abstract. Supports conditional inconsistency under specified evolutionary regimes, not a claim about every dataset. registry ↩a ↩b ↩c
[3] Pasi Hyytiäinen, "The Changing Text of Acts: A Phylogenetic Approach", TC: A Journal of Biblical Textual Criticism 26 (2021), PDF pp. 13–18. Directly supports the 54-witness/279-variation-place matrix and MP application. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g