From Few-Body Inputs to Many-Body Predictions

Few-body data determine a many-body observable only after every symmetry-allowed scale at the target accuracy has been included. The scattering length fixes a leading two-body amplitude; effective range, resonance width, three-body data, temperature, confinement, loss, and medium operators can supply independent corrections. A defensible prediction propagates them together and stops when the first uncontrolled contribution reaches the claimed precision.

Required background. Universal Relations and Tan Contact supplies nonperturbative cross-checks, and Few-Body Data in the Virial Expansion supplies a controlled few-to-many-body expansion.

Helpful background. Resonant Bose Matter and Metastable Branches shows how lifetime can bound the physical interpretation.

Dimensionless statement of the prediction

Choose an observable $O$ and its natural units $O_{\rm ref}$ . For a homogeneous three-dimensional gas, write schematically

\frac{O}{O_{\rm ref}} =F\!\left( k_Fa,\,k_Fr_e,\,k_FR^*,\, \frac{k_F}{\kappa_*},\, \frac{T}{T_F},\,z,\, k_Fa_\perp,\, \frac{t}{t_n},\,\frac{t}{\tau_3},\ldots \right).

Only the arguments relevant to the species, geometry, state, and method are retained. Writing them before calculating prevents the unitary limit $a^{-1}=0$ from being mistaken for the absence of all microscopic input.

The momentum $k_F$ here can be replaced by $k_n$ , $\lambda_T^{-1}$ , a binding momentum, or a probe scale. Use the largest momentum actually resolved when estimating derivative-operator corrections.

Four sources of uncertainty

Input uncertainty. Experimental determinations of $a$ , $r_e$ , $R^*$ , and $\kappa_*$ can be correlated. If the parameter vector is $\boldsymbol\theta$ with covariance $C_\theta$ , linear propagation gives

\sigma_{\mathrm{input}}^2 =J_i(C_\theta)_{ij}J_j, \qquad J_i=\frac{\partial O}{\partial\theta_i}.

Near a pole, linear propagation can fail; sample the full input likelihood instead.

Theory truncation. If terms through order $Q^\nu$ are retained, estimate the first omitted order from its power counting and coefficients inferred from lower orders or regulator-consistent variants. For example, a zero-range calculation should be varied by an $r_e$ term and then by the leading shape term. The EFT uncertainty framework and its assumptions are reviewed by Furnstahl, Phillips, and Wesolowski 2015.

Numerical uncertainty. Vary momentum and frequency cutoffs after renormalization, discretization, basis, volume, continuation method, and solver tolerances. A regulator variation is a truncation diagnostic only after the couplings have been rematched at each regulator.

State and measurement uncertainty. Include trap averaging, density calibration, finite preparation time, loss, heating, probe matrix elements, and final-state interactions. These are not absorbed into the few-body scattering error.

When a new many-body input appears

Few-body vacuum matching fixes local operators already present in the low-energy action. A medium can make additional structures leading through:

a Fermi surface or condensate that changes infrared power counting;
an order parameter and stiffness not fixed by vacuum scattering alone;
long-range Coulomb or dipolar forces;
lattice band parameters and broken Galilean symmetry;
a three-body counterterm promoted by renormalization;
dissipative coefficients after deep channels are eliminated; or
probe-specific vertices not determined by the equation of state.

The appearance of such an input is not a failure of universality. It defines the correct, less restrictive universality class.

Worked uncertainty example

Suppose a fermionic equation of state at unitarity is calculated through linear effective range,

\frac{E}{N}=\frac35E_F \left[\xi+\zeta_e k_F r_e+O\!\left((k_F r_e)^2,k_F R\right)\right].

Take $k_Fr_e=0.08\pm0.01$ , $k_FR=0.02$ , $\zeta_e=0.12\pm0.03$ , and a numerical error $0.003$ on the bracket. The range shift is $0.0096$ . Its input-and-coefficient uncertainty is

\sigma_{r_e} =\sqrt{(0.12\times0.01)^2+(0.08\times0.03)^2} \simeq0.0027.

If natural coefficients are assumed, the next effective-range term is about $(0.08)^2\simeq0.0064$ and the microscopic-range correction about $0.02$ . The latter dominates; combining independent contributions in quadrature gives roughly $0.021$ before state-systematic errors. A sub-percent claim would therefore be unsupported even though the fitted input error is small.

Correlated theory errors should not be combined in quadrature blindly. If two variations probe the same omitted operator, treat them as one source; if a sign or coefficient is unknown, report the interval or probability model used.

Cross-checks that change the conclusion

Use at least one constraint not fitted by the calculation:

compare contact from a tail with contact from an adiabatic derivative;
compare compressibility from the equation of state with static response;
verify a virial limit at small $z$ ;
recover a known weak-coupling coefficient;
compare broad- and two-channel calculations as $k_FR^*\to0$ ;
repeat a metastable observable at several hold times and densities.

Failure of an independent check is evidence that the uncertainty model omitted a contribution. Enlarging a fit error without identifying the missing structure does not repair the prediction.

Stopping rule

State a numerical conclusion only if:

the matched result is regulator independent at the retained order;
each expansion parameter is evaluated over the complete kinematic window;
input, truncation, numerical, and state uncertainties are reported with correlations;
at least one held-out exact limit or observable agrees; and
the state lifetime supports the equilibrium or transient language used.

Otherwise report a qualitative trend or an interval restricted to the tested regime. The conclusion should never be stronger than the least controlled item above.

Common pitfalls

Propagating only the error on $a$ . Near resonance, range, width, and three-body uncertainty often dominate because $a^{-1}$ is deliberately small.

Varying a cutoff without rematching. That changes the physical two-body input and confounds regulator dependence with a different theory.

Adding correlated errors as independent. Several scheme variations can represent the same omitted term.

Exercises

Propagate correlated inputs

For $O=Aa+Br_e$ , derive its input variance when $\operatorname{Cov}(a,r_e)=\rho\sigma_a\sigma_r$ .

Solution

The Jacobian is $(A,B)$ , so $\sigma_O^2=A^2\sigma_a^2+B^2\sigma_r^2+2AB\rho\sigma_a\sigma_r$ . Positive correlation increases the error when $A$ and $B$ have the same sign and decreases it when their contributions oppose.

Apply the stopping rule

A calculation is cutoff stable and satisfies the contact relation, but $t/\tau_3=0.7$ during data acquisition and loss heating is unmodeled. What claim survives?

Solution

The formal zero-range calculation may describe a fixed-density Hamiltonian observable, but comparison with the measured state lacks a controlled state correction. One may report a model trend or an early-time comparison; an equilibrium equation-of-state extraction is not supported.

Continue

Resonant Interactions and Universal Quantum Gases summarizes the hierarchy. Cooper Instability and Pairing applies the matched interaction to a new infrared state. Baym–Kadanoff Conservation and Validity provides complementary response and approximation checks.

References

Furnstahl, R. J., D. R. Phillips, and S. Wesolowski. “A Recipe for EFT Uncertainty Quantification in Nuclear Physics.” Journal of Physics G: Nuclear and Particle Physics 42 (2015): 034028. DOI.

From Few-Body Inputs to Many-Body Predictions

Dimensionless statement of the prediction

Four sources of uncertainty

When a new many-body input appears

Worked uncertainty example

Cross-checks that change the conclusion

Stopping rule

Common pitfalls

Exercises

Propagate correlated inputs

Apply the stopping rule

Continue

References

Further reading