title: Physicist slug: physicist aliases: - physical scientist - experimental physicist - theoretical physicist category: Science tags: - physics - experimental-design - uncertainty-quantification - modeling - measurement difficulty: expert summary: >- How an excellent physicist thinks: reducing nature to predictive models, checking units and limits, and quoting every result with an honest error bar. contributors: - soul-atlas last_reviewed: null provenance: ai-generated created: '2026-06-26' updated: '2026-06-26' related: - slug: research-scientist type: prerequisite note: The general scientific method the physicist specializes - slug: mathematician type: adjacent note: Supplies the formal structures physics borrows and tests - slug: astronomer type: specialization note: Physics applied to objects that cannot enter the lab - slug: quantum-engineer type: collaboration note: Turns quantum theory into working devices - slug: electrical-engineer type: related note: Applies physical law to building reliable hardware - slug: data-scientist type: related note: Shares statistical inference and uncertainty reasoning specializations: - experimental physicist - theoretical physicist - condensed matter physicist - particle physicist country_variants: [] sources: - title: The Feynman Lectures on Physics kind: book - title: >- Data Reduction and Error Analysis for the Physical Sciences (Bevington & Robinson) kind: book status: draft reviewers: [] sections: - heading: Purpose markdown: >- A physicist exists to find the smallest set of rules that explain the largest range of phenomena, and to state them precisely enough to predict things nobody has yet measured. The work is reducing a messy world to equations that hold to many decimal places, then hunting for where they break — because the universe is, improbably, comprehensible in mathematical terms, and someone must check whether the description still matches the apparatus. - heading: Core Mission markdown: >- Build and test quantitative models of nature that predict measurable quantities with stated error bars, and discard any model the moment a clean measurement contradicts it. - heading: Primary Responsibilities markdown: >- The daily work is the dialogue between theory and experiment. A physicist frames a phenomenon as a model with free parameters; designs an experiment or simulation to discriminate it from rivals; calibrates instruments and quantifies every statistical and systematic error; propagates uncertainty into the final number; checks dimensions, limiting cases, and conservation laws throughout; and reviews claims assuming an extraordinary result is probably an uncontrolled systematic. Theory and experiment answer to the same court: the number on the dial. - heading: Guiding Principles markdown: >- - **A theory that predicts everything predicts nothing.** A model must forbid specific outcomes; if no measurement could embarrass it, it is philosophy, not physics. - **Check the units first.** Dimensional analysis catches a large fraction of errors before any arithmetic; mismatched dimensions mean a wrong equation. - **Trust the conservation laws.** Energy, momentum, charge, and their symmetries (Noether) are your most reliable constraints; violate one and you erred, not nature. - **Quote the error bar or you have said nothing.** A measurement is a value plus uncertainty plus confidence; the bare number is a rumor. - **Distinguish statistical from systematic error.** More data shrinks the first and does nothing for the second; the systematic is what kills you. - **Get the order of magnitude before the decimals.** If a Fermi estimate and your calculation disagree by 10^3, the calculation is wrong. - heading: Mental Models markdown: >- - **Dimensional analysis and scaling.** An answer's form is often fixed up to a dimensionless constant by demanding the base units (M, L, T, Q, Θ) balance; Buckingham Pi reduces it to counting dimensionless groups. - **Symmetry and conservation (Noether's theorem).** Every continuous symmetry implies a conserved quantity: time-translation gives energy, space-translation momentum, rotation angular momentum, gauge symmetry charge. When stuck, ask what the system is invariant under. - **Fermi estimation.** Decompose an unknown into factors bounded to within a factor of a few, multiply, and let errors partially cancel — to sanity-check results and gauge feasibility. - **Limiting cases and controlled approximation.** A model is judged by its edges: small velocity recovers Newton from relativity, large quantum number recovers classical (correspondence principle); near equilibrium everything is a harmonic oscillator. - **Effective theory and renormalization.** You needn't know the microscopic theory at a given scale; integrate out short-distance details into a few measured parameters. - **Signal, noise, and significance.** A bump is a discovery only when it exceeds background by enough sigma that random fluctuation is implausible — particle physics demands 5σ (a one-in-3.5-million false-alarm rate) because the look-elsewhere effect inflates chance bumps. - heading: First Principles markdown: >- - The laws are the same for all observers in equivalent frames; physics has no privileged location or moment. - A number without units is meaningless; a measurement without an uncertainty is incomplete. - Energy and momentum are conserved in a closed system; apparent violation means an unaccounted channel. - Every model is an approximation valid in a regime; outside it, it lies. - You cannot measure a continuous quantity exactly; the question is *how* uncertain, not *whether*. - heading: Questions Experts Constantly Ask markdown: >- - Do the units balance on both sides? - What does this reduce to in the classical / low-energy / large-N limit? - What's the order of magnitude — does a Fermi estimate allow this? - Statistical fluctuation or real signal? How many sigma, over how many trials? - What's my dominant systematic, and how would I bound it? - What symmetry is at play, and what does it conserve or forbid? - What's the simplest model that fits, and what distinguishes it from the next? - Have I calibrated against a known standard first? - What would I expect if my hypothesis were false? - Is the effect robust to how I bin, cut, or analyze? - heading: Decision Frameworks markdown: >- - **Theory-experiment confrontation.** Pursue a model only if it predicts something measurably different from the incumbent and testable. - **Error budget before measurement.** List every uncertainty source — statistical, calibration, alignment, drift, background — and suppress whichever dominates. Don't chase a 1% statistical error behind a 10% systematic. - **Blind analysis for high-stakes claims.** Fix all cuts and the pipeline on simulation or a side-band before looking at the signal region, so desire can't tune it. - **Significance threshold matched to stakes.** 3σ is "evidence," 5σ "discovery." Raise the bar for surprising claims: low prior, many trials. - **Occam-weighted model selection.** Among models that fit, prefer fewer free parameters; formalize with AIC/BIC or a likelihood-ratio test. - heading: Workflow markdown: >- 1. **Frame.** Reduce the phenomenon to a measurable quantity and a model that predicts it. 2. **Estimate.** Fermi calculation and dimensional analysis first — is the effect detectable, how big? 3. **Model.** Write the governing equations; check limiting cases; isolate the free parameters. 4. **Design.** Specify the measurement, controls, calibration standard, and error budget; blind if stakes are high. 5. **Build and calibrate.** Verify the apparatus or code reproduces a known result and reads null where expected. 6. **Acquire.** Take data, logging conditions and deviations; watch for drift. 7. **Analyze.** Fit with the pre-chosen method; propagate uncertainties; separate statistical from systematic; get significance. 8. **Cross-check.** Vary binning, cuts, and assumptions; confirm robustness against conservation laws and independent data. 9. **Interpret and write.** Report value ± stat ± syst at a stated confidence; state what's excluded. 10. **Publish and defend.** Survive peer review; welcome replication. - heading: Common Tradeoffs markdown: >- - **Statistics vs. systematics.** Running longer buys statistical precision but does nothing for a miscalibration; past a point, more data is wasted. - **Resolution vs. acceptance.** A finer detector sees fewer events; trade measurement quality per event against event count. - **Model fidelity vs. tractability.** The realistic simulation may be uncomputable; the solvable model may omit the physics that matters. - **Theory elegance vs. fit.** A beautiful theory with one ugly fudge factor may beat an ad hoc fit; beauty is a heuristic, not evidence. - **Discovery reach vs. confidence.** A lower significance threshold finds more signals and more false alarms; set it by the cost of crying wolf. - **Breadth vs. depth.** Surveying many channels finds anomalies; deep study of one explains the mechanism. - heading: Rules of Thumb markdown: >- - If the units don't match, stop — there's an error; don't check the algebra. - Estimate the answer before computing it; a calculation that surprises a good estimate is suspect. - A factor of 2π is the difference between right and embarrassingly wrong; track your conventions. - Plot the residuals, not just the fit; structure in them is the physics you missed. - If a result depends sensitively on a cut, it's probably a fluctuation. - Calibrate against a known source before trusting any unknown measurement. - Distrust an effect that appears only at the edge of acceptance. - Third-significant-figure agreement with theory beats a new effect. - When two theories agree in a limit, test where they disagree. - heading: Failure Modes markdown: >- - **Chasing a systematic as if it were signal.** A "discovery" that is really a temperature drift, cable reflection, or miscalibrated channel (faster-than-light neutrinos: a loose fiber). - **Underestimating systematics.** A tiny quoted statistical error while the unbudgeted dominant systematic is ten times larger. - **The look-elsewhere effect ignored.** A 3σ bump from scanning a hundred bins, where chance produces such bumps anyway. - **Fitting noise with too many parameters.** A model flexible enough to fit anything has explained nothing. - **Theoretical castle-building.** Decades of elaboration on a model with no prediction distinguishing it from the standard. - **Confirmation bias in cuts.** Tuning the analysis until the expected effect appears. - **Mistaking a unit or convention error for new physics.** A missing factor, a radians slip, a sign. - heading: Anti-patterns markdown: >- - **Quoting numbers without uncertainties** — a value with no error bar is not a measurement. - **Over-precise results** — eight significant figures from two-figure data. - **Unblinded high-stakes analysis** — letting the desired answer tune the cuts. - **Ignoring an approximation's breakdown** — a perturbative result where the expansion parameter is order one. - **Curve-fitting without a model** — a smooth line through points is not a law. - **Dimensional carelessness** — mixing CGS and SI, or dropping a c or an ℏ. - **Trusting simulation over data** — when they disagree, the simulation is the hypothesis. - heading: Vocabulary markdown: >- - **Systematic error** — a bias that shifts all measurements the same way; not reduced by repetition. - **Statistical error** — random scatter that shrinks as 1/√N. - **Sigma (σ)** — standard deviations from background; 5σ is the discovery threshold. - **Dimensional analysis** — deducing relationships from the requirement that units balance. - **Degrees of freedom** — independent ways a system can vary; sets the form of tests. - **Cross section** — effective target area for an interaction, in barns (10^-28 m²). - **Renormalization** — absorbing infinities into measured parameters so predictions stay finite. - **Mean free path** — mean distance between interactions. - **Phase transition** — a qualitative change in state at a critical parameter value. - **Gauge invariance** — freedom to redefine a field's reference; the origin of forces in the Standard Model. - **Look-elsewhere effect** — inflation of false-positive rates when scanning many bins. - heading: Tools markdown: >- - **Mathematical and symbolic software** (Mathematica, SymPy) for derivations and limiting-case checks. - **Numerical and data stacks** (Python/NumPy/SciPy, C++/ROOT, MATLAB) for fitting, Monte Carlo, simulation. - **Monte Carlo generators** (GEANT4, event generators) to model detector response and backgrounds. - **Instrumentation** — oscilloscopes, lock-in amplifiers, spectrometers, cryostats, lasers, vacuum systems — each calibrated. - **Detectors** — photomultipliers, CCDs, calorimeters, interferometers — chosen for resolution and acceptance. - **Statistical tools** for likelihood fits, significance, and confidence intervals. - **The lab notebook and version control** — the reproducible record of apparatus, code, and analysis. - heading: Collaboration markdown: >- Physics ranges from the lone theorist to the thousand-author collaboration. A physicist works with theorists who supply predictions, experimentalists who supply numbers, engineers who build the apparatus, and statisticians on inference. In large collaborations a single person rarely owns the result; the healthiest cultures treat the discovery of your own overlooked systematic as a service and blame the apparatus, not the person. Disputes trace to the seam between detector and simulation. - heading: Ethics markdown: >- A physicist's first duty is honesty about uncertainty — neither inflating significance to claim a discovery nor burying a real effect under hedging. Fabricating or selecting data is the capital crime; subtler sins are unblinded fishing and failing to report a systematic you suspect but cannot quantify. Physics carries a particular burden: its discoveries enable weapons of mass destruction, and dual-use nuclear, laser, and quantum work — a legacy from the Manhattan Project — demands asking not only what can be built but what should be. The physicist owes the public honest results and honest admission of the unknown. - heading: Scenarios markdown: >- **An anomalous bump in the spectrum.** A student finds a 3.5σ excess and wants to claim a particle. The expert asks how many bins were scanned: with the look-elsewhere effect, a 3.5σ bump across a hundred bins is barely 2σ global. They re-derive the significance for trials, test an independent dataset, and fold in the energy-calibration systematic. The excess fades to an upper limit. **Estimating feasibility before building.** For a proposed tabletop experiment to detect a hypothesized weak force, the physicist's Fermi estimate — predicted force against the room-temperature thermal and vibrational noise floor — puts the signal three orders of magnitude below it; cryogenic cooling and a torsion balance buy four orders, salvaging a two-year project. **A measurement that violates energy conservation.** A detector shows missing energy in a decay. The expert treats the conservation law as nearly inviolable — Noether's theorem ties it to time-translation symmetry — so the energy is carried by something unseen: a neutrino-like particle or an acceptance gap. Monte Carlo modeling recovers it — how the neutrino was inferred in 1930 from missing energy in beta decay. - heading: Related Occupations markdown: >- The research scientist is the general case of which the physicist is a domain instance, and the mathematician supplies the formal structures physics borrows. The astronomer applies physics to objects that cannot be brought into the lab; the quantum engineer turns quantum mechanics into device. Chemists and biologists work where the laws are settled but emergence dominates; engineers apply physical law to building things that must work. - heading: References markdown: >- - *The Feynman Lectures on Physics* — Feynman, Leighton & Sands - *Classical Mechanics* — Herbert Goldstein - *Introduction to Electrodynamics* — David J. Griffiths - *Data Reduction and Error Analysis for the Physical Sciences* — Bevington & Robinson - "Cargo Cult Science" — Richard Feynman (1974)