REVIEW 5 major objections 5 minor 2 cited by
An Improvement of AmpRed: Analytic Continuation of Complex Integrals
T0 review · 5 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash
Pith's one-line read Choosing an auxiliary scale with sign-definite coefficients lets AmpRed evaluate complex integrals as fast as real ones, according to benchmarks up to four loops.
desk verdict A modest but real software improvement for evaluating complex multi-loop integrals; the branch-fixing rule is plausible but the 'always possible' claim needs qualification. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central object is the auxiliary scale insertion $y=x_i/x_j$ through $\delta(y-E^{(0)}(x))$, which rewrites the integral as $\int dy\, y^{\lambda_i} I_y$ with $F_y=F|_{x_i=y x_j}$. The mechanism that makes complex evaluation cheap is sign-definiteness: if the coefficients of the monomials in $F_y$ are all positive (or all negative), the $-i0^+$ prescription is claimed to fix the branch of $I_y$, so the subsequent differential-equation computation runs as for real kinematics. The paper notes that this condition is always satisfiable for loop integrals because the coefficient of $x_{n+1}$ in $F$ equals the positive-definite first Symanzik polynomial $U$.
What would settle it
Take a multi-scale integral at a complex kinematic point where the physical sheet is known independently, for example the two-loop example in Fig. 1a, and evaluate it both with a sign-definite auxiliary choice and with AMFlow's independent method. If the values disagree or land on different sheets for any sign-definite choice, the claim that sign-definiteness fixes the branch is falsified.
Extended reading notes
Core claim
The central claim is that the physical Riemann sheet of the auxiliary integral $I_y$ is fixed by the $-i0^+$ prescription whenever all coefficients of the monomials in $F_y$ have the same sign. For ordinary loop integrals such a pair $\{x_i,x_j\}$ always exists, because the coefficient of $x_{n+1}$ in $F$ is the first Symanzik polynomial $U$, which is positive definite. The paper therefore asserts that complex master integrals can be computed with the same algorithm and nearly the same cost as real integrals, without the expensive branch-point search of the earlier method. The claim is supported by timing benchmarks rather than by a proof of the branch-fixing observation.
Load-bearing premise
The entire speedup rests on an unproved observation: when the auxiliary variable is chosen so that all coefficients in the integrand's expansion have the same sign, the integration automatically picks the physically correct branch. If that observation fails for some integral, the method would return a value on the wrong Riemann sheet.
Editorial extensions
If this is right
- AmpRed can now evaluate complex master integrals at essentially the same cost as real ones, removing the main practical penalty of complex kinematics.
- On the three- and four-loop benchmarks, the new implementation is about three to four times faster than AMFlow, and it handled the Wilson-line integral that made AMFlow run out of memory, using about 200 GB of RAM.
- The method applies to integrals with linear propagators (Wilson lines) as well as ordinary loop integrals, because the sign-definite auxiliary choice relies only on the positive definiteness of the first Symanzik polynomial.
- For multi-scale integrals, the paper recommends using differential equations with the boundary condition chosen at a singular point, together with the scaling relation $I(-s)=e^{-i\pi(\lambda_0+\lambda_{n+1}+1)}I(s)^*$ for real $s$.
Reading between the lines
- If the branch-fixing observation is correct, the same sign-definiteness criterion should apply to cut integrals and to auxiliary variables other than $x_i/x_j$; the paper explicitly leaves cut integrals out, so testing those cases would extend the method.
- The reported speedups depend on the hardware, the IBP solver, the precision goal, and the Euclidean-region choice; the qualitative advantage is likely to persist, but the exact factors are setup-specific.
- A rigorous proof of the branch-fixing statement would upgrade the empirical benchmarks into a theorem; a targeted search for a counterexample in higher-loop multi-scale integrals would test where the condition starts to fail.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. This short paper reports an update to the AmpRed package for evaluating multi-loop Feynman master integrals in the Feynman-parameter representation. The proposed improvement targets the analytic continuation of complex integrals: by choosing the auxiliary scale pair {x_i, x_j} in the insertion E^(0) = x_i/x_j so that the coefficients of the monomials y^a in F_y have definite signs, the author claims that the branch of I_y is fixed by the -i0+ prescription of F. This would remove the need for the slower branch-fixing procedure of the previous version (ref. [5], sec. 3.4) and make complex integrals almost as efficient as real ones. The paper presents benchmarks for four integrals, from two to four loops, comparing AmpRed with AMFlow, and reports that AmpRed is faster for the three- and four-loop examples while AMFlow fails on the Wilson-line integral due to memory constraints. The main methodological claim, however, is stated as an observation without proof, and the paper does not provide a direct comparison with the previous AmpRed version or explicitly report numerical agreement with AMFlow.
Significance. If the branch-fixing observation is correct, the method offers a practically valuable improvement for evaluating master integrals at complex kinematic points, a common need in multi-loop calculations. The paper provides a concrete algorithmic change, a reproducible example notebook, and benchmarks against an independent package, which are all positives. The central claim is falsifiable and can be tested on the provided examples. However, the unproved branch-fixing rule is load-bearing: if it fails, the computed integrals lie on the wrong Riemann sheet and the timing comparisons are meaningless. The absence of a direct comparison with the previous AmpRed implementation and the lack of an explicit statement of numerical agreement with AMFlow further weaken the paper's headline claims.
major comments (5)
- [Paragraph after Eq. (5)] The branch-fixing rule is stated as an 'observation' without proof or reference. The claim that the branch of I_y is fixed by the -i0+ prescription of F whenever the coefficients of the monomials y^a in F_y are positive (or negative) definite is the load-bearing step of the paper; if the rule fails, the computed integrals sit on the wrong Riemann sheet and the benchmark timings do not establish the method's correctness. Please provide a proof or a rigorous analytic-continuation argument, or give a precise reference to a derivation.
- [Paragraph after Eq. (5)] The statement that such a choice of {x_i,x_j} is 'always possible for loop integrals' is not supported by the given justification. Choosing i = n+1 makes the coefficient of the linear monomial x_{n+1} in F equal to the first Symanzik polynomial U, which is positive definite, but after the substitution x_{n+1} = y x_j the y^0 coefficient of F_y is the second Symanzik polynomial F, which is not sign-definite in general. The later comment that the benchmarks are performed in the Euclidean region, where F has more positive terms, does not establish strict sign-definiteness of all monomial coefficients in F_y. Please state the precise condition needed and verify that it holds for each benchmark integral.
- [Abstract and Table I] The abstract claims that the new implementation 'significantly enhances computational efficiency' compared to the previous version of AmpRed, but Table I compares AmpRed only against AMFlow, not against the previous AmpRed method (e.g., the sec. 3.4 algorithm of ref. [5]). Without a direct timing comparison to the old implementation, the abstract's comparative claim is unsupported. Please add such a comparison or revise the abstract to claim only the efficiency relative to AMFlow.
- [Table I and surrounding text] The paper does not state whether the numerical values produced by AmpRed and AMFlow agree. Since the method's central purpose is to obtain complex master integrals on the correct branch, the timing comparison is meaningful only if both packages yield the same results to the stated precision goal. Please report the numerical agreement explicitly or provide the computed values, e.g., as supplementary material.
- [Text after Table I] The statement that 'the efficiency is thus close to that of real integrals' is not benchmarked. No comparison is shown between AmpRed's run time for a complex kinematic point and its run time for a real kinematic point on the same integral. Either provide such a comparison or qualify the claim so that it refers only to what Table I actually demonstrates.
minor comments (5)
- [Eq. (5)] The notation 'Fλ0 y' is ambiguous; it should be written as F_y^{λ0} to clearly denote the power of F_y.
- [Eq. (4)] The transformation of the integration measure from dΠ(n+1) to dΠ(n) after integrating out the δ-function is not explained; a short comment would help the reader follow the derivation.
- [Table I] The line 'Precision goal: 20' is a fragment; please state it as a full sentence, e.g., 'All integrals were computed to 20 significant digits' or 'The precision goal was 20 digits for both packages.'
- [References] Reference [5] omits the publication year and arXiv identifier; please complete the citation.
- [Fig. 1 caption] For reproducibility, the caption or text should state the specific Euclidean kinematic point (e.g., values of Mandelstam variables and masses) used in the benchmarks, rather than referring only to the example notebook.
Circularity Check
No significant circularity: the improved analytic-continuation method is an algorithmic claim benchmarked externally, not a derivation that reduces to its own inputs.
full rationale
I walked the derivation chain of the paper. The central algorithmic claim is that for complex integrals one can choose the auxiliary scale pair {xi, xj} so that the coefficients of monomials in Fy are sign-definite, which then fixes the branch through the -i0+ prescription. This is presented as a mathematical observation about the Feynman-parameter representation and the Symanzik polynomials, not as a conclusion presupposed by the method. The selection rule is justified by the independent property that the coefficient of xn+1 in F is the first Symanzik polynomial U, which is positive definite for loop integrals; this property is an input about the integrand structure, not about the numerical value of the master integral being predicted. The efficiency claim is supported by benchmark computations in Table I comparing AmpRed with the external package AMFlow on the same integrals, with no fitted parameter that is later renamed as a prediction. The paper does cite the author's prior work for the underlying parametric-reduction and iterative algorithms, but the specific improvement under discussion is not established by citing those works; it is instead demonstrated by the stated branch-fixing reasoning and by the external timing comparison. The main weakness is that the branch-fixing observation after Eq. (5) is asserted rather than proved, and the accompanying 'always possible' justification addresses only the xn+1 coefficient while the y^0 coefficient can be the second Symanzik polynomial F, whose sign-definiteness is not guaranteed in general. That is a correctness or rigor concern, not circularity: a failed or incomplete proof does not make the derivation equivalent to its input by construction. Since no step in the paper reduces a predicted quantity to a fitted input, to a self-citation chain, or to a definition, the appropriate circularity score is 0.
Assumptions & free parameters
assumptions (4)
- ad hoc to paper The branch of Iy is fixed by the -i0+ prescription of F when the coefficients of monomials ya in Fy are positive (or negative) definite.
- ad hoc to paper Such a choice of pair {xi,xj} is always possible for loop integrals because the coefficient of x_{n+1}, namely the first Symanzik polynomial U, is always positive definite.
- domain assumption The Euclidean-region relation I(-s)=e^{-iπ(λ0+λn+1+1)} I(s)^* extends to the integrals computed.
- domain assumption The parametric Feynman integral representation (1) and the iterative algorithm from ref. [5] are valid.
Cite this review
Pith. "Pith review of An Improvement of AmpRed: Analytic Continuation of Complex Integrals." pith.science (2026). https://pith.science/paper/HWBQFJWW
@misc{pith2026250513540,
author = {Pith},
title = {Pith review of: An Improvement of AmpRed: Analytic Continuation of Complex Integrals},
year = {2026},
howpublished = {\url{https://pith.science/paper/HWBQFJWW}},
note = {Machine review of arXiv:2505.13540}
}
read the original abstract
The AmpRed package has been updated with an improved method for analytic continuation of complex integrals. Compared to the previous version, the new implementation significantly enhances computational efficiency for evaluating complex integrals.
Figures
Forward citations
Cited by 2 Pith papers
-
Spacelike-Collinear Scattering by the Method of Regions
The kinematic factorisation-violating part of the two-loop spacelike-collinear splitting amplitude comes entirely from a single hidden region with soft and Glauber loop momenta.
-
AMFlow 2.0: significant algorithmic and software improvements for Feynman integral evaluation
AMFlow 2.0 cuts symbolic and numerical cost of multi-loop Feynman integral evaluation via an FT recursion mode, a C++ DE solver, and modern IBP reducers, demonstrated on a three-loop five-point family.
Reference graph
Works this paper leans on
-
[5]
W. Chen, M.-x. Luo, T.-Z. Yang, and H. X. Zhu, Soft theorem to three loops in QCD and N = 4 super Yang-Mills theory, JHEP 01, 131, arXiv:2309.03832 [hep-ph]
-
[1]
Chen, Semi-automatic calculations of multi-loop Feynman amplitudes with AmpRed, Comput
W. Chen, Semi-automatic calculations of multi-loop Feynman amplitudes with AmpRed, Comput. Phys. Commun. 312, 109607 (2025), arXiv:2408.06426 [hep-ph]
arXiv 2025
-
[2]
W. Chen, Reduction of Feynman Integrals in the Parametric Representation, JHEP 02, 115, arXiv:1902.10387 [hep-ph]
arXiv 1902
-
[3]
W. Chen, Reduction of Feynman Integrals in the Parametric Representation II: Reduction of Tensor Integrals, Eur. Phys. J. C 81, 244 (2021), arXiv:1912.08606 [hep-ph]
arXiv 2021
-
[4]
Chen, Reduction of Feynman integrals in the parametric representation III: integrals with cuts, Eur
W. Chen, Reduction of Feynman integrals in the parametric representation III: integrals with cuts, Eur. Phys. J. C 80, 1173 (2020), arXiv:2007.00507 [hep-ph]
arXiv 2020
-
[6]
F. V. Tkachov, A theorem on analytical calculability of 4-loop renormalization group functions, Phys. Lett. B 100, 65 (1981)
1981
-
[7]
K. G. Chetyrkin and F. V. Tkachov, Integration by parts: The algorithm to calculate β-functions in 4 loops, Nucl. Phys. B 192, 159 (1981)
1981
-
[8]
Laporta, High-precision calculation of multiloop Feynman integrals by difference equations, Int
S. Laporta, High-precision calculation of multiloop Feynman integrals by difference equations, Int. J. Mod. Phys. A 15, 5087 (2000), arXiv:hep-ph/0102033
arXiv 2000
Show all 15 references
-
[9]
A. V. Kotikov, Differential equations method: New technique for massive Feynman diagrams calculation, Phys. Lett. B 254, 158 (1991)
1991
-
[10]
Remiddi, Differential equations for Feynman graph amplitudes, Nuovo Cim
E. Remiddi, Differential equations for Feynman graph amplitudes, Nuovo Cim. A 110, 1435 (1997), arXiv:hep-th/9711188
1997 arXiv
-
[11]
Gehrmann and E
T. Gehrmann and E. Remiddi, Differential equations for two-loop four-point functions., Nucl. Phys. B 580, 485 (2000), arXiv:hep-ph/9912329
2000 arXiv
-
[12]
Liu and Y.-Q
X. Liu and Y.-Q. Ma, AMFlow: A Mathematica package for Feynman integrals computation via auxiliary mass flow, Comput. Phys. Commun. 283, 108565 (2023), arXiv:2201.11669 [hep-ph]
2023 arXiv
-
[13]
Maierh¨ ofer, J
P. Maierh¨ ofer, J. Usovitsch, and P. Uwer, Kira—A Feynman integral reduction program, Comput. Phys. Commun. 230, 99 (2018), arXiv:1705.05610 [hep-ph]
2018 arXiv
-
[14]
Kauers, Fast Solvers for Dense Linear Systems, Nucl
M. Kauers, Fast Solvers for Dense Linear Systems, Nucl. Phys. B Proc. Suppl. 183, 245 (2008)
2008
-
[15]
Klappert, F
J. Klappert, F. Lange, P. Maierh¨ ofer, and J. Usovitsch, Integral reduction with Kira 2.0 and finite field methods, Comput. Phys. Commun. 266, 108024 (2021), arXiv:2008.06494 [hep-ph]
2021 arXiv
Reviewed August 15, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.