AgentSkillsCN

vector-awkward

在 Python 中使用 scikit-hep 向量和 Awkward Array 构建向量记录、注册行为、计算 deltaR 或不变质量、组合或增强向量,或从 ak.zip 记录中访问向量属性(Momentum3D/Momentum4D/Vector4D 类型)。只有同时使用 Awkward,向量才有意义。

SKILL.md
--- frontmatter
name: vector-awkward
description: Use when working in Python with scikit-hep vector and Awkward Array to build vector records, register behaviors, compute deltaR or invariant masses, combine or boost vectors, or access vector properties from ak.zip records of type (Momentum3D/Momentum4D/Vector4D). Vector is only useful if you are also using Awkward.

Vector Awkward

Overview

Create Awkward Arrays that behave like vector objects and use vector methods without writing custom kinematic math. Use these steps to register behaviors, build records with the right field names, and perform common HEP operations like deltaR and invariant mass.

Quick Start

  1. Register vector behaviors once per session:
python
import vector
vector.register_awkward()
  1. Build vector records with standard field names and a vector type name:
python
import awkward as ak

events = ak.Array({
    "electron": ak.zip(
        {"pt": [50.0, 30.2], "eta": [1.4, -0.8], "phi": [2.1, 0.5], "mass": [0.0005, 0.0005]},
        with_name="Momentum4D",
    )
})

Build Vector Records

  • Use ak.zip(..., with_name="Momentum3D") for 3D operations like deltaR.
  • Use ak.zip(..., with_name="Momentum4D") for 4D operations like invariant mass or boosts.
  • Prefer standard names like pt/eta/phi/mass or px/py/pz/E so vector can infer coordinate systems.
  • Call vector.register_awkward() before accessing vector properties or methods.

Access Vector Properties

python
particles = ak.zip({"px": px, "py": py, "pz": pz, "E": E}, with_name="Vector4D")
particles.pt
particles.phi
particles.eta
particles.mass

Common Operations

Calculate deltaR between two collections:

python
pairs = ak.cartesian([events.electron, events.muon])
electrons, muons = ak.unzip(pairs)
dR = electrons.deltaR(muons)

Combine 4-vectors and compute invariant mass:

python
first_e, second_e = ak.unzip(ak.combinations(events.electron, 2))
inv_mass = (first_e + second_e).mass

Boost to a parent rest frame:

python
parent = particle1 + particle2
particle1_rf = particle1.boostCM_of_p4(parent)

Reference Material

Load references/vector-hints.md for more examples, method notes, and advanced snippets.