Theoretical foundations and applications of the Loop-Tree Duality in Quantum Field Theories

The thing worth your time here is Chapter 7 and 8: an actual C++ program that evaluates dual integrals numerically, with a three-dimensional contour deformation, checked against LoopTools for scalar integrals through the pentagon and tensor integrals through the hexagon. The theory chapters are recaps of the author's published papers [1,2,48,49] and are presented as such; the new content is the implementation and the deformation. That is a genuine enabling step for the LTD program, and the thesis is honest about which parts are borrowed and which are new. The singularity classification in §7.3–7.4 — hyperboloid pairs cancel among dual contributions, ellipsoids need deformation, coupled contributions must share the same deformation — is clean and useful, and the grouping strategy in Fig. 7.4 is the kind of practical detail that saves the next person months. I checked the sign condition κ·qi<0 in §7.5.2; it survives the full sum over pairs, not just term by term, though the thesis doesn't spell that out. The stress-test concern about the deformation is fair and lands. The only guard against the contour enclosing an unwanted pole is the λ̄ rescaling of Eq. (7.23), which solves for one on-shell energy at a time. Overlapping ellipsoids within a deformation group, and the kji,0→0 limit where the dual prescription degenerates, are not covered by that argument. §7.6 concedes the chosen λ=-0.5, A=10⁶ give 'very good results for most of the examples', which implies some points don't. The empirical LoopTools comparison is the right test, but the tables are truncated in what I could read, so I can't confirm the agreement or the error budget. Two smaller things: the code isn't public, and §5.2.1 states outright that no formal proof exists that IBP always removes multiple poles — the author flags this himself, so it's a known gap, not a hidden one. Also, everything validated is finite; IR/UV cases are explicitly out of scope (footnote 1, §7.5.2). None of this is fatal. It's a thesis doing a first implementation, and it does it. Worth a referee, mainly to force a proper error budget and a public code release.

The central claim is the first numerical LTD implementation, validated against LoopTools. For that to hold, the deformed contour must (a) carry the correct i0 sign at every dual pole, (b) vanish at infinity, and (c) not enclose any other pole. Conditions (a) and (b) are argued in §7.5.2: Eq. (7.16) requires κ·qi<0, and the exponential in Eq. (7.18) localizes the deformation. I checked that (a) survives the full sum over pairs, not just term by term: v_ij·qi = |qi| + qj·qi/|qj| ≥ 0, so with λij<0 every term contributes κ·qi ≤ 0. The paper does not spell this out, but the step is correct. Condition (c) rests entirely on Eq. (7.23), which solves q(+)i,0(𝓁+iλ̄κ)=0 for one on-shell energy at a time and caps λ̄ at (1/2)√Y or √(X²−Y/4). That is a single-pole analysis. It does not address configurations where two ellipsoid singularities of the same deformation group overlap in loop three-momentum space, so that the summed deformation of Eq. (7.18) at one pole is dominated by a different pair than the one being guarded. Nor does it address the limit kji,0→0, where the dual prescription −i0 kji,0 degenerates and the required deformation direction becomes ambiguous. If the contour crosses a dual pole, the result differs from LoopTools by a residue, so the empirical comparison is the right test. But the supplied text truncates the tables in §7.7.2–§8.5, so the agreement cannot be verified here, and §7.6 concedes 'very good results for most of the examples', implying some points fail.

The reader's weakest_assumption points at the contour deformation, which is the right place to look, but frames it as 'hand-chosen parameters'. That framing is slightly off target: λ and A in Eq. (7.18) control accuracy and efficiency, not correctness, and the paper's own λ-scan in §7.6.1–7.6.2 shows the integral is stable across a range of them. The genuinely load-bearing question is narrower: whether the deformed contour stays on the correct side of every dual pole for arbitrary kinematics. The paper's construction is more defensible than the reader allows. Condition (a) of §7.5.2 holds for the full summed deformation, as I verified above, and condition (b) follows from the exponential suppression. The residual soft spot is condition (c), guarded only by the heuristic λ̄ rescaling of Eq. (7.23), which is a one-pole-at-a-time analysis. That is a real gap, but it is bounded: a crossed pole would produce a residue-sized discrepancy against LoopTools, and the paper does run that comparison. So the empirical check is the correct one, and the concern lands only if the comparison fails at some sampled point. The supplied manuscript truncates the numerical tables, so I cannot confirm or refute the agreement here. The paper also acknowledges several limitations that should not be re-raised as new findings: §7.5.2 footnote 1 states that soft and collinear singularities are out of scope because 'we are only dealing with finite diagrams'; §5.2.1 states 'Although no formal proof exists' for the IBP reduction to single poles; and the code is not released. These are honest disclosures, not hidden assumptions. The thesis is a compilation of published work ([1], [2]) plus a new implementation, and the published chapters are supported by their citations. On the central claim, the method is coherent and the validation strategy is appropriate; the verdict CONDITIONAL is the right level, and my read does not move it. I would only add that the abstract's 'for the first time' is consistent with the group's own earlier proceedings [5,6], which are self-cited and do not undercut the claim. One editorial note, not load-bearing: Table 8.4 in the hexagon section repeats the caption of Table 8.3 ('Examples for pentagons involving numerators of tensor rank two and three'), which looks like a copy-paste slip and should be corrected before the results are read as hexagon-level evidence.

If the numerical claim holds, the thesis is a useful proof of concept: it shows that dual integrals, which are tree-level objects over a modified phase space, can actually be integrated numerically for multi-leg one-loop configurations, which is the prerequisite for the programme of combining virtual and real corrections in a single Monte Carlo generator. Several things deserve explicit credit. The validation is against LoopTools, an independent implementation, so the central numerical check is not circular. The integrand-level analysis of Ch. 6, in particular the demonstration in Eqs. (6.11)-(6.17) that singularities from intersecting forward hyperboloids cancel between dual contributions, is a genuine structural result and is what makes the dual representation numerically attractive in the first place. The construction of the deformation in Eq. (7.18), with a vector part that orients the shift away from the ellipsoid and a Gaussian suppression that localizes it, is concrete and implementable rather than schematic. The thesis is also candid about scope: it states that only finite diagrams are treated and that the reduction of multiple poles to single poles is not proved in general

This thesis sets out to show that a virtual one-loop Feynman integral can be traded for a sum of single-cut, tree-level objects integrated over a three-dimensional phase space, and — the part that is new here — that this rewriting can actually be run as a computer program for multi-leg kinematics. The route is the residue theorem applied to the loop-energy integral: each residue puts one internal line on shell and turns the remaining propagators into 'dual' propagators carrying a modified $i0$ prescription, giving $L^{(1)}=-~L^{(1)}$ with no double or higher cuts. The author then pushes the formalism outward in three directions: to two loops, to propagators raised to higher powers (double and triple poles, reduced away with integration-by-parts identities), and to a study of how singularities are shared and cancelled between dual contributions and between virtual and real corrections. The concrete deliverable is a C++ program built on the Cuba integrators that deforms the loop three-momentum into the complex plane to step around the surviving poles, and whose output is compared against LoopTools for scalar triangles, boxes and pentagons and for tensor numerators up to hexagon level. If the approach holds up, the practical payoff is that loop and real corrections stop being separate computational objects and could be handled inside one Monte Carlo event generator.

What carries the argument. The load-bearing object is the dual propagator $G_D(q_i;q_j)=1/(q_j^2-m_j^2-i0~k_{ji,0})$, obtained by evaluating the other Feynman propagators at the residue of the cut line, with $k_{ji}=q_j-q_i$ fixed by external momenta alone. Its virtue is that it can be rewritten as $~L^{(1)}=-~L^{(1)}$-form $~L^{(1)}=-~L^{(1)}$, so that each dual integral is a genuine three-dimensional integral over the on-shell forward hyperboloid of one internal line, with the energy already fixed. Two singularity types follow from this form: the ellipsoid ($k_{ji}^2-(m_j+m_i)^2>0$, $k_{ji,0}<0$) and the hyperboloid ($k_{ji}^2-(m_j-m_i)^2<0$). The second piece of machinery is the contour deformation $~L^{(1)}=-~L^{(1)}$ with $i~L^{(1)}=-~L^{(1)}$, a sum over ellipsoids of a vector part pointing outward from the singularity surface times a Gaussian-like suppression $~L^{(1)}=-~L^{(1)}$, whose scale $~L^{(1)}=-~L^{(1)}$ and width $~L^{(1)}=-~L^{(1)}$ are tuned per contribution and then globally rescaled by a safety factor derived from requiring that the deformed on-shell energy not cross another pole.

Core claim. The central claim is a first: the Loop-Tree Duality has been implemented as a working numerical program that computes one-loop scattering diagrams. Concretely, the thesis derives the dual representation $L^{(1)}(p_1,...,p_N)=-~L^{(1)}(p_1,...,p_N)$ with $~L^{(1)}=-i$-normalised single-cut integrals $~L^{(1)}=-~L^{(1)}$ built from $~L^{(1)}=-~L^{(1)}$-type terms, then shows how to organise the numerical evaluation: map the infinite loop three-momentum onto the unit cube in spherical coordinates, classify each dual propagator's singularity as a forward-forward hyperboloid (which cancels between dual contributions) or a forward-backward ellipsoid (which must be dodged), group the dual contributions that share hyperboloids so they receive an identical deformation, and integrate with Cuhre or VEGAS. The program is validated point-by-point and by scans against LoopTools, including scans through thresholds and through the region where a deformation becomes mandatory and the integrand turns complex. Alongside this, the thesis extends the formalism itself: a Lorentz-invariant residue formula for double poles, an integration-by-parts strategy that removes multiple poles before duality is applied at two and three loops, and an argument that singularities from intersecting forward light-cones cancel at the integrand level while the survivors are confined to a finite region of loop momentum of the order of the external momenta — which is what makes a local map onto the real-correction phase space conceivable.

The premise it rests on. The numerical agreement rests on a contour deformation whose shape and scale ($~L^{(1)}=-~L^{(1)}$) were chosen by hand and checked only on the momentum configurations that were tested, so the claim that the program computes one-loop diagrams generally depends on that deformation avoiding every pole and leaving the integral unchanged for kinematics nobody sampled; separately, the author acknowledges that no formal proof exists that integration-by-parts reduction always eliminates multiple poles, only that it did so in every case studied.

Dual integrals are three-dimensional rather than four-dimensional, so the numerical cost of a one-loop virtual correction scales like a phase-space integral rather than a loop integral.

Because forward-forward hyperboloid singularities cancel between dual contributions at the integrand level, a large class of integrable singularities needs no contour deformation at all.

Multiple poles, unavoidable at two loops, can be removed by integration-by-parts reduction without reducing to a master-integral basis, leaving the single-pole duality relation applicable unchanged.

The surviving singularities are confined to loop momenta of the order of the external momenta, which is the precondition for mapping the virtual region onto the real-correction phase space and cancelling soft and collinear divergences locally.

A working dual-integral evaluator makes it plausible to combine loop and tree contributions inside a single Monte Carlo generator instead of subtracting them against each other.

The natural next test is a process where the dual and real integrands are actually summed before integration rather than compared afterwards; the thesis argues the cancellation is local but does not demonstrate it numerically.

Since the deformation parameters are hand-tuned per contribution and then rescaled, an adaptive scheme that reads the singularity structure off the external momenta would remove the main source of fragility and is the obvious engineering step.

The tensor results up to hexagon level suggest the method is not blocked by numerator algebra, which matters because integration-by-parts reduction of multiple poles generates exactly such numerators.

The framework's real target — NLO cross sections with many external legs — would require the dual integrand to be built from full scattering amplitudes rather than scalar integrals, including gauge-dependent numerator terms the thesis explicitly sets aside.

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review