Physical foundations for mathematicians

Choose this pathway when you understand a formal QFT framework but want to know what physical problem its objects represent, how physicists construct a calculation, and what evidence can support or contradict the result. The route connects actions, states, correlators, particles, amplitudes, response functions, approximations, and measurements without treating any one of them as the definition of QFT.

The intended output is a comparison between one formal statement and one physical construction: what is prepared, what is varied or probed, what is calculated, what is observed, which approximation is used, and where the interpretation stops.

Required background. Linear maps, duals, and basic Hilbert-space quantum mechanics. If operator domains, spectra, states, or pictures are unfamiliar, begin with quantum states and operators. The route supplies its own bridges through variational mechanics, relativity, and field quantization.

Begin with an experiment-shaped question

A mathematical structure becomes physically interpretable only after a preparation and a readout are specified. Start with this template:

system and regime:
state preparation or boundary data:
source, interaction, or intervention:
quantity calculated:
measurement or operational interpretation:
approximations and limiting procedure:
independent check or evidence:

For example: prepare two widely separated wave packets of a stable massive scalar, let them scatter in the vacuum at center-of-mass energy $\sqrt s$, calculate the leading connected $2\to2$ amplitude from a quartic interaction, and interpret the corresponding cross section as an idealized event rate. The statement already assumes isolated one-particle poles, an in/out regime, a vacuum background, perturbative coupling, and a measurement that can identify the final particles.

This habit prevents three common substitutions:

• a field symbol is not automatically a particle or a measured quantity;
• a correlation function is not automatically a probability or cross section; and
• a rigorous object is not automatically the object an experiment prepares.

Follow the common physical spine

Use the route in three passes.

1. Build the classical and kinematic model

Variational and classical fields explains how an action, admissible variations, boundary terms, constraints, and equations of motion fit together. Relativity, Lorentz symmetry, and spin separates transformation laws of fields from the classification of particle states.

Then use Classical fields, actions, and local dynamics to derive one field equation and retain the boundary assumptions. An action is compact dynamical data; it is not yet a quantum theory, a state, or an observable.

2. Distinguish the quantum objects

Quantum fields, states, and observables supplies the central dictionary. Continue through canonical quantization of the free scalar and functional integrals and correlators to compare operator and source-based constructions in the one case where both can be calculated exactly.

Do not demand an ordinary particle interpretation in every setting. A particle is tied to a representation, state, and asymptotic or spectral structure. Local fields and algebras can remain meaningful in thermal states, curved spacetime, confined theories, or critical systems where a vacuum S-matrix is not the right observable.

3. Add interactions, consistency, and scale

Symmetry, currents, and Ward identities shows how an invariance constrains correlators only after contact, boundary, breaking, and possible anomaly terms are included. Renormalization and the RG separates the regulator used to define an intermediate calculation from the finite inputs and physical scale dependence of the result.

At that point, Structural principles and axiom maps and Theorem-first frameworks and claim grammar can be read as maps from physical requirements—causality, stability, positivity, locality, symmetry, and state preparation—to precise mathematical hypotheses.

Running bridge: a scalar action becomes a rate

Consider a real scalar in four-dimensional Minkowski spacetime,

$$\mathcal L =\frac12\partial_\mu\phi\,\partial^\mu\phi -\frac12m^2\phi^2 -\frac{\lambda}{4!}\phi^4, \qquad m>0.$$

Varying the action gives

$$\delta S =-\int_\Omega\mathrm d^4x\, \left(\Box\phi+m^2\phi+\frac{\lambda}{3!}\phi^3\right)\delta\phi +\int_{\partial\Omega}\mathrm d\Sigma_\mu\, \partial^\mu\phi\,\delta\phi.$$

The Euler–Lagrange equation follows only after the allowed variations or boundary condition make the surface term vanish. This is the first physical choice: the same bulk formula can belong to different boundary-value problems.

When $\lambda=0$, canonical quantization promotes the smeared field to an operator-valued distribution and supplies a vacuum and one-particle states with

$$\langle\boldsymbol p'\vert\boldsymbol p\rangle =(2\pi)^3,2E_{\boldsymbol p} \delta^{(3)}(\boldsymbol p'-\boldsymbol p).$$

The two-point function has an isolated pole at $p^2=m^2$. That pole, its positive residue, and the state normalization—not the letter $\phi$—justify using the field as an interpolating operator for a stable particle.

Turn on $\lambda$ perturbatively. The quartic term gives the tree vertex $-i\lambda$. LSZ reduction removes the external one-particle poles and yields

$$\mathcal M(\phi\phi\to\phi\phi)=-\lambda.$$

For equal-mass elastic scattering, full solid-angle integration, and two identical final particles,

$$\frac{\mathrm d\sigma}{\mathrm d\Omega} =\frac{\lambda^2}{128\pi^2s}, \qquad \sigma=\frac{\lambda^2}{32\pi s}.$$

The action did not itself contain a cross section. The chain also used a vacuum, stable asymptotic states, field normalization, a perturbative truncation, flux, phase space, and an identical-particle convention. The original reduction theorem makes the stable-pole and asymptotic assumptions explicit Lehmann, Symanzik, and Zimmermann 1955, pp. 205–225.

This result has clear evidence and failure tests: it has cross-section dimension $-2$; it is invariant under exchange of identical scalars; a contact amplitude has no exchange pole; loop corrections must obey unitarity and renormalization-scale consistency; and ordinary LSZ fails if the external excitation is unstable, confined, or lacks an isolated pole.

Keep the meanings of “observable” separate

Object	Physical job	What it is not by itself
Self-adjoint operator or POVM	Encodes possible outcomes for a specified quantum measurement	A detector model or a local relativistic observable in every representation
Local observable algebra	Organizes operations available in a spacetime region	A preferred field coordinate, vacuum, or particle basis
Time-ordered correlator	Generates perturbative responses and pole data	A directly normalized probability or rate
Retarded correlator	Describes linear response to a source in a chosen state	An in/out scattering amplitude
S-matrix amplitude	Encodes an idealized transition between asymptotic states	A cross section before flux, phase space, and measurement weights
Cross section or response spectrum	Connects theory to a class of event counts or probe outcomes	Raw detector data without acceptance, resolution, and inference

The local-algebra usage originates in the assignment of observable algebras to regions Haag and Kastler 1964, pp. 848–861. The scattering usage is appropriate only when in/out states exist. In curved spacetime or accelerated motion, a localized detector response can be meaningful even when no global particle number is preferred; Unruh’s detector analysis is a classic example Unruh 1976, pp. 870–892.

Separate idealization, approximation, and evidence

A physical calculation usually has several logically different layers:

1. Model idealization: field content, background, symmetry, boundary conditions, and state.
2. Mathematical definition: smearing, regulator, finite volume, gauge fixing, or operator domain.
3. Approximation: perturbative order, semiclassical expansion, EFT power counting, saddle point, or discretization.
4. Numerical estimator: quadrature, sampling, truncation, solver tolerance, and statistical uncertainty.
5. Measurement interface: source and detector model, binning, calibration, acceptance, and inference.

Do not use success at one layer as proof of another. A converged numerical integral supports the discretized calculation, not the physical model’s applicability. Agreement with data at one point tests the complete pipeline there, not every intermediate interpretation. A regulator may disappear from a renormalized prediction; a physical cutoff or EFT breakdown scale cannot be removed by notation.

A good comparison names one independent check per layer: symmetry or conservation for the model, regulator independence for the definition, order-by-order residuals for the approximation, convergence and benchmark tests for the numerics, and held-out or calibrated checks for the measurement interface.

Choose a branch by the physical question

Conformal correlators and bootstrap data

Choose correlators, OPE, and conformal blocks when the formal input is conformal symmetry and the physical questions concern critical observables, spectra of scaling dimensions, operator products, or response near a fixed point. Track how Euclidean correlators, radial quantization, reflection positivity, crossing, and experimental or lattice critical exponents enter. A numerical exclusion region is evidence conditional on symmetry, unitarity, truncation, and solver control—not a measurement on its own.

Fields on curved spacetime

Choose fields and local algebras on curved backgrounds when geometry and state selection replace global translation invariance. Identify the background, causal propagator, algebra or field, Hadamard state condition, and local observable. Particle creation and detector response depend on states, observers, and asymptotic regimes; do not assume a unique vacuum or ordinary S-matrix.

Dualities and global data

Choose field-theory duality: dictionaries, operations, and global data when two descriptions are claimed to encode the same physics. Build a dictionary for operators, states, symmetries, parameters, extended objects, and observables; then test protected quantities, anomalies, spectra, partition functions, or controlled limits. Similar low-energy equations are not sufficient evidence for a full duality, and global forms or line operators can distinguish theories with the same Lie algebra.

Produce a framework-to-physics comparison

End with one compact record:

formal statement and object class:
physical preparation and probe:
quantity calculated and operational interpretation:
idealizations and approximation order:
normalization and convention choices:
two independent checks:
evidence currently available:
claim that remains unsupported:

For the scalar example, the formal object is the connected four-point function with stable one-particle poles; the preparation is two incoming wave packets; LSZ supplies the amplitude; flux and phase space supply the rate; tree order is the approximation; exchange symmetry and mass dimension are checks; and no claim has yet been made about loop accuracy, detector response, or high-energy validity.

Exercises

1. Keep the boundary term

Derive the scalar variation above. State two different boundary choices that make the variational problem differentiable and explain why the bulk equation alone does not choose between them.

Solution

The kinetic variation is

$$\int_\Omega\mathrm d^4x\, \partial_\mu\phi\,\partial^\mu\delta\phi =-\int_\Omega\mathrm d^4x\, (\Box\phi)\delta\phi +\int_{\partial\Omega}\mathrm d\Sigma_\mu\, \partial^\mu\phi\,\delta\phi.$$

Adding the potential variation gives the displayed result. Dirichlet data fix $\phi$ on the boundary and require $\delta\phi|_{\partial\Omega}=0$. A Neumann problem can instead fix the normal derivative after adding the appropriate boundary term or choosing data that make the surface variation vanish. Both lead to the same local bulk equation while defining different admissible solutions and physical boundary data.

2. Diagnose a category error

A calculation produces a time-ordered two-point function with a pole. Its author calls the function “the probability of detecting a particle.” List the additional steps and hypotheses needed for that interpretation.

Solution

First establish that the pole is isolated, has the correct positive residue, and belongs to a physical stable state rather than a gauge-dependent or confined field. A time-ordered correlator supplies propagation and spectral information but is not positive pointwise and is not normalized as a probability. To predict scattering, specify in/out wave packets, apply LSZ, and then combine the amplitude with flux, phase space, state sums, and a measurement definition. To predict a localized detector, specify the detector coupling, trajectory, switching, state, and response function instead. The two operational questions need not agree outside a common asymptotic regime.

3. Classify the error sources

A lattice calculation of a response function agrees with an experiment in one bin. Sort the following into layers: finite lattice spacing, Monte Carlo autocorrelation, omitted operators in an EFT, detector migration, and a wrong symmetry assignment. Give one check for each.

Solution

• Finite lattice spacing belongs to mathematical definition and discretization; vary the spacing and extrapolate with the expected scaling law.
• Monte Carlo autocorrelation belongs to the numerical estimator; estimate the integrated autocorrelation time and verify blocked or replicated uncertainties.
• Omitted EFT operators belong to approximation; compare successive orders and test residual scaling with the declared small parameter.
• Detector migration belongs to the measurement interface; validate the response matrix with calibration or held-out simulation and propagate its covariance.
• A wrong symmetry assignment belongs to the model idealization; test the corresponding Ward identity, selection rule, or transformation of measured channels.

Agreement in one bin does not identify which layer is correct; several errors can compensate there.

References

• Haag, Rudolf, and Daniel Kastler. “An Algebraic Approach to Quantum Field Theory.” Journal of Mathematical Physics 5 (1964): 848–861. DOI.
• Lehmann, Harry, Kurt Symanzik, and Wolfhart Zimmermann. “On the Formulation of Quantized Field Theories.” Il Nuovo Cimento 1 (1955): 205–225. DOI.
• Unruh, William G. “Notes on Black-Hole Evaporation.” Physical Review D 14 (1976): 870–892. DOI.

Continue with one branch above and complete the comparison template for a real question. If your next task is instead to state a formal claim and its hypotheses more precisely, use Mathematical foundations for physicists. If you need a sequential physics course, begin with QFT I.