Mathematical foundations for physicists

Choose this pathway when the difficult part of your question is no longer a Feynman integral but the status of the objects being used. It helps a physicist turn statements such as “the Euclidean theory continues to a unitary QFT,” “local observables commute,” or “this interacting model exists” into claims with explicit hypotheses, object classes, conclusions, and failure boundaries.

The route does not replace physical intuition or calculation with abstraction. Its output is a precise map between one physical claim and the framework in which that claim can actually be stated and tested.

Required background. You should be comfortable with quantum states and operators and with Lorentz transformations and causal separation. If either is unstable, use the focused reviews of quantum states and operators or relativity, Lorentz symmetry, and spin. Helpful background. Distribution theory, complex analysis, correlators, Ward identities, and RG reasoning become necessary in different branches; the route points to them when used.

Translate a physics claim into a theorem-shaped question

Begin with one sentence from your actual work. Rewrite it under six headings:

physical claim:
mathematical objects and their regularity:
hypotheses, including dimension and background:
conclusion and its quantifiers:
notion of equivalence or reconstruction:
known failure boundary:

For example, “the Euclidean path integral defines the Lorentzian theory” is too compressed. A usable version asks whether a specified hierarchy of Euclidean Schwinger distributions satisfies Euclidean covariance, permutation symmetry, reflection positivity, regularity, and clustering strongly enough for an Osterwalder–Schrader reconstruction. The conclusion then concerns a Hilbert space, vacuum, positive energy, and Lorentzian fields—not the pointwise existence of an oscillatory functional measure.

Three distinctions prevent most category errors:

• Objects: fields as operator-valued distributions, local observable algebras, Euclidean measures, time-ordered products, and cochain complexes are not interchangeable inputs.
• Result class: a reconstruction theorem, a structural theorem, a perturbative construction, and a nonperturbative existence proof answer different questions.
• Scope: spacetime dimension, background geometry, state class, ultraviolet cutoff, interaction, and topology can be hypotheses rather than incidental details.

Open Theorem-first frameworks and claim grammar once you have written the six-line version. Use it to name the framework before searching for a theorem.

Build only the common language you need

The mathematical reviews at the beginning of this route are tools, not a qualifying examination:

1. Linear and tensor methods clarifies maps, duals, adjoints, representations, and invariant statements.
2. Fourier transforms, distributions, and Green functions supplies test-function topology, weak equations, support, and boundary-value prescriptions.
3. Complex and asymptotic methods separates analytic continuation from formal substitution and controlled asymptotics from convergent equality.

Skip a review if you can already perform the operation that your claim uses. Then align the physics:

• Fields, states, and observables separates the principal object classes.
• Functional integrals and correlators distinguishes a regulated construction from a formal measure and connects Gaussian Euclidean and Lorentzian kernels.
• Symmetry, currents, and Ward identities keeps contact terms, boundaries, explicit breaking, and Jacobians visible.
• Renormalization and the RG distinguishes a regulator, a continuum limit, a renormalization scheme, and physical scale dependence.

This common spine is enough when the goal is to compare hypotheses. It is not a demand to master every formalism before choosing a branch.

Worked bridge: reflection positivity for a free scalar

The Euclidean two-point function of a free scalar of mass $m>0$ can be written

$$S_2(x-y) =\int\frac{\mathrm d^d p}{(2\pi)^d} \frac{e^{ip\cdot(x-y)}}{p^2+m^2}.$$

Let $x=(\tau,\boldsymbol x)$ and reflect Euclidean time by $\theta x=(-\tau,\boldsymbol x)$. Performing the frequency integral gives

$$S_2(\tau,\boldsymbol x) =\int\frac{\mathrm d^{d-1}\boldsymbol p}{(2\pi)^{d-1}} \frac{e^{i\boldsymbol p\cdot\boldsymbol x} e^{-E_{\boldsymbol p}|\tau|}}{2E_{\boldsymbol p}}, \qquad E_{\boldsymbol p}=\sqrt{\boldsymbol p^2+m^2}.$$

For a smooth test function $f$ supported at positive Euclidean time, consider

$$Q[f] =\int_{\tau_x,\tau_y>0}\mathrm d^d x\,\mathrm d^d y\, \overline{f(x)}\,S_2(\theta x-y)\,f(y).$$

Because $|{-\tau_x-\tau_y}|=\tau_x+\tau_y$, the expression factorizes:

$$Q[f] =\int\frac{\mathrm d^{d-1}\boldsymbol p}{(2\pi)^{d-1}} \frac{1}{2E_{\boldsymbol p}} \left\lvert \int_{\tau>0}\mathrm d\tau\,\mathrm d^{d-1}\boldsymbol x\, e^{-E_{\boldsymbol p}\tau-i\boldsymbol p\cdot\boldsymbol x} f(\tau,\boldsymbol x) \right\rvert^2 \geq0.$$

This is reflection positivity for the free covariance. It is the Euclidean trace of Hilbert-space positivity and positive energy: time translation away from the reflection plane contributes $e^{-E\tau}$ with $E\geq0$. The original reconstruction program states the necessary conditions for a complete hierarchy of Euclidean Green functions in Osterwalder and Schrader 1973, pp. 83–112.

The calculation proves less than a common slogan suggests. Positivity of this one two-point function does not construct an interacting theory. A full reconstruction uses all $n$-point Schwinger distributions and additional covariance, symmetry, regularity, and cluster hypotheses. Nor does analytic continuation mean replacing $\tau$ by $it$ in an arbitrary numerical function; the continuation is controlled by the reconstructed spectral and distributional structure.

Choose the branch that matches the claim

Vacuum fields and Wightman reconstruction

Use Wightman fields, reconstruction, and structural theorems when the input is a hierarchy of Lorentzian vacuum expectation distributions and the question concerns positive energy, locality, covariance, CPT, spin–statistics, or reconstruction of fields and a Hilbert space. The classic formulation treats fields as operator-valued tempered distributions Gårding and Wightman 1956, pp. 860–866.

This branch exposes domains and distributional smearing that textbook notation often suppresses. It does not, by axioms alone, prove that a chosen interacting four-dimensional model exists.

Euclidean fields and Osterwalder–Schrader reconstruction

Use Euclidean fields, reflection positivity, and reconstruction when the input is Euclidean Schwinger data or a Euclidean measure and the goal is a positive-energy Lorentzian theory. Your work product should identify the reflection, positive-time test space, positivity form, regularity class, and continuation target.

This is the natural branch for lattice-to-continuum and constructive questions, but a discretization or regulator must preserve enough positivity and symmetry for the intended limit. A converged Euclidean numerical integral is not by itself a reconstruction theorem.

Local algebras and locally covariant QFT

Use local nets, states, and representations when the primary objects are algebras assigned to spacetime regions. Isotony, spacelike commutativity, covariance, and the chosen state or representation replace a commitment to one privileged field coordinate. This language is particularly effective for superselection sectors, inequivalent representations, and curved spacetime; the local-net program begins with Haag and Kastler 1964, pp. 848–861.

The branch does not automatically supply a Lagrangian, a preferred vacuum on a general background, or an easy particle interpretation. State selection remains physical data.

Constructive QFT and cutoff removal

Use constructive Euclidean QFT and cutoff removal when the claim is that an interacting model exists nonperturbatively and satisfies specified axioms after ultraviolet and volume cutoffs are removed. The decisive objects are uniform bounds, tightness or convergence, correlation inequalities, and control of the limiting Schwinger functions.

Results are strongly model- and dimension-dependent. Existence of selected two- and three-dimensional models does not license an existence claim for an arbitrary four-dimensional theory.

Microlocal QFT and curved backgrounds

Use microlocal QFT and renormalized local fields when singular directions, Hadamard states, products of distributions, time-ordered products, or local covariance on curved spacetime are central. Wavefront sets refine the vague instruction to “avoid coincident singularities,” and the microlocal spectrum condition replaces global Fourier support when translation symmetry is absent Brunetti, Fredenhagen, and Köhler 1996, pp. 633–652.

Local renormalized fields do not create a preferred global vacuum or S-matrix. Geometry, state, and global causal structure still bound the claim.

Factorization algebras and local-to-global observables

Use factorization algebras and local-to-global observables when observables are organized cohomologically and disjoint regions combine through a local-to-global multiplication. This is especially natural for perturbative BV constructions and for comparisons with topology and geometry.

A factorization algebra and a Haag–Kastler net package locality differently. Any claimed equivalence must state the categories, analytic hypotheses, quantum versus classical setting, and whether the comparison is perturbative. Similar vocabulary does not establish identity.

Produce a hypothesis map

End the route with one page that another reader can challenge line by line:

Field	Free-scalar reflection-positivity example
Physical claim	The Euclidean free covariance is compatible with a positive-energy Lorentzian scalar field
Input object	A tempered Euclidean two-point distribution with mass $m>0$
Key hypothesis	Positive-time support for test functions and time reflection $\theta(\tau,\boldsymbol x)=(-\tau,\boldsymbol x)$
Decisive check	$Q[f]$ is an integral of $
Result class	One reflection-positivity check contributing to Euclidean reconstruction
Not established	An interacting model, the full Schwinger hierarchy, or a cutoff-removal theorem
Next branch	Euclidean reconstruction for the theorem; constructive QFT if existence after cutoff removal is the claim

For a new problem, replace every cell. A citation belongs beside the precise theorem or construction it supports; it cannot substitute for naming the hypotheses.

Common pitfalls

Axioms are a construction. Axioms define a class of theories and enable structural theorems. Showing that a nontrivial interacting model belongs to that class is a separate existence problem.

Analytic continuation is a symbol replacement. Continuation requires an analytic domain or reconstruction theorem and control of boundary values as distributions. Numerical smoothness on the Euclidean axis is insufficient.

All frameworks have the same observables. Fields, bounded local algebras, Euclidean random variables, and cochain observables may be related under additional hypotheses, but their equality is not built into their names.

Rigor removes physical choices. A mathematically complete result can still depend on a state, representation, boundary condition, background, or renormalization prescription. Precision makes those choices visible; it does not choose them automatically.

Exercises

1. Reproduce the positivity factorization

Starting from the mixed representation of $S_2$, derive the expression for $Q[f]$ as an integral of an absolute square. Identify exactly where $m>0$, positive-time support, and positive energy enter.

Solution

For $\tau_x,\tau_y>0$,

$$S_2(\theta x-y) =\int\frac{\mathrm d^{d-1}\boldsymbol p}{(2\pi)^{d-1}} \frac{e^{-E_{\boldsymbol p}(\tau_x+\tau_y)} e^{i\boldsymbol p\cdot(\boldsymbol x-\boldsymbol y)}}{2E_{\boldsymbol p}}.$$

Insert this into $Q[f]$, interchange the test-function and momentum integrals, and group the $x$ and $y$ factors as complex conjugates. The result is the absolute-square formula on the page. Positive-time support turns the reflected separation into $\tau_x+\tau_y$; $E_{\boldsymbol p}>0$ makes both the exponential and $1/(2E_{\boldsymbol p})$ positive. The assumption $m>0$ avoids a zero-energy singularity at $\boldsymbol p=0$; massless fields require an additional infrared-domain check, although reflection positivity can still hold on an appropriate test space.

2. Repair an overclaim

A simulation finds a stable, positive Euclidean two-point function as the lattice spacing decreases. The report concludes: “The interacting Lorentzian QFT has been constructed.” Rewrite the conclusion at the strongest level supported by the stated evidence and list the missing steps.

Solution

The supported claim is narrower: the measured two-point observable appears numerically stable under the tested lattice refinements and is compatible with positivity in the tested configurations. To claim a reconstructed interacting QFT, one would need a specified continuum limit for the full relevant Schwinger hierarchy, reflection positivity as a functional inequality, Euclidean covariance or its recovered continuum form, symmetry and regularity conditions, clustering when required, and control of infinite-volume and renormalized limits. Numerical systematics and the state or phase selected by the simulation must also be stated. The observation is useful evidence toward those steps, not their completion.

3. Translate microcausality between frameworks

State the local commutativity condition for smeared Wightman fields and the corresponding Haag–Kastler condition for local observable algebras. Why is this a translation rather than a proof of equivalence?

Solution

For bosonic fields, if the supports of test functions $f$ and $g$ are spacelike separated, Wightman locality requires

$$[\phi(f),\phi(g)]=0$$

on a common invariant domain. A local net assigns algebras $\mathcal A(O)$ to regions and requires

$$[A,B]=0 \quad\text{for all }A\in\mathcal A(O_1),\ B\in\mathcal A(O_2)$$

when $O_1$ and $O_2$ are spacelike separated. The two statements express causal compatibility in different object classes. Passing from fields to a net requires defining suitable bounded observables or generated algebras and controlling domains; reconstructing fields from a net requires additional assumptions. Locality alone does not provide either construction.

References

• Brunetti, Romeo, Klaus Fredenhagen, and Michael Köhler. “The Microlocal Spectrum Condition and Wick Polynomials of Free Fields on Curved Spacetimes.” Communications in Mathematical Physics 180 (1996): 633–652. DOI.
• Gårding, Lars, and Arthur S. Wightman. “Fields as Operator-Valued Distributions in Relativistic Quantum Theory.” Physical Review 101 (1956): 860–866. DOI.
• Haag, Rudolf, and Daniel Kastler. “An Algebraic Approach to Quantum Field Theory.” Journal of Mathematical Physics 5 (1964): 848–861. DOI.
• Osterwalder, Konrad, and Robert Schrader. “Axioms for Euclidean Green’s Functions.” Communications in Mathematical Physics 31 (1973): 83–112. DOI.

Continue with exactly one framework branch above. If your real need is the reverse translation—from formal structures to actions, states, approximations, and measurements—use Physical foundations for mathematicians. If the claim is already precise and the next obstacle is a calculation, return to Learning pathways and choose by that calculation.