{"id":"57953adc-4ed0-43dd-84f2-1b045acd67d2","arxiv_id":"2506.19766","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A modular framework speeds up energy-momentum tensor calculations from transport models by clustering particles into representative super-particles while keeping averaged physics accurate.","lead":"This paper introduces MATRICS, a software framework that computes the smooth energy and charge distributions used in heavy-ion collision simulations. It clusters particles into representative super-particles and adapts grids to speed up calculations, which could help large-scale studies of dense QCD matter.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Representative particles drop intra-cluster momentum spread, so the spatial stress tensor T^ij is not preserved; \"perfect agreement\" is validated only on energy density, baryon density, and flow.","rationale":"The reader's weakest assumption was that a single representative particle reproduces the smeared tensor at every grid point, citing the nonlinearity of Eq. (3). My stress test sharpens this into a concrete failure mode: the spatial stress tensor is not approximately preserved when intra-cluster momenta cancel, and in a locally thermalized cluster that cancellation is exactly the thermal pressure. The paper's quantitative checks are weighted transverse-plane averages that are insensitive to T^ij, so the demonstration does not cover the full tensor claimed in the abstract and Sec. 4.3. This is a load-bearing concern for applications to hydrodynamic initial conditions, but the paper may still be valid for the restricted set of observables it actually plots (energy density, baryon density, flow, and geometric anisotropies). The appropriate response is to require the additional validation or an explicit narrowing of the claim, which matches the existing CONDITIONAL verdict; hence I recommend no change to the reader's verdict.","tokens_in":16557,"tokens_out":8576,"duration_ms":99956,"concrete_test":"Add a panel to the Fig. 3 comparison showing the local isotropic pressure P = (T^xx + T^yy + T^zz)/3 in the Landau rest frame, and one off-diagonal component such as T^xy, at z = 0 on the identical grid, for \"No cluster + Smear\" versus \"Box cluster (dyn) + Smear\". Also run the analytic two-particle test: with equal-mass particles of momenta +p and -p in one cluster, the original T^xx equals 2 p_x^2 / E while the representative gives 0. If the full-event pressure or shear components differ by more than a few percent in cells that carry most of the energy, the \"perfect agreement\" claim fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that a representative particle defined by Eqs. (1)-(2) preserves the energy-momentum tensor of Eq. (3). This cannot hold for the spatial components, because Eq. (3) is quadratic in momentum. Consider a cluster of two identical particles at the same point with momenta +p and -p. The original tensor has T^xx = 2 p_x^2 / E, while the representative has P = (2E, 0), giving T^xx_rep = 0. In general, the representative replaces sum_i p_i^j p_i^k / p_i^0 with P^j P^k / P^0; the missing term is the momentum-space variance of the cluster, which in a locally thermalized cell is precisely the pressure and shear stress. The paper's validations in Fig. 3 are transverse-plane weighted averages of e, n_B, transverse flow, and eccentricities; these are dominated by T^00, T^0i, and the energy-density geometry, so they do not test T^ij. The nonlinearity is acknowledged in Sec. 4.1.3, but the comparison there also does not report spatial stress. Thus the Sec. 4.3 claim of \"perfect agreement\" with the baseline is not established for the full energy-momentum tensor, and the method is not safe for hydrodynamic initialization or pressure-gradient-sensitive observables unless T^ij is separately validated or the representative carries additional moments.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper describes MATRICS, a workflow for computing space-time distributions of the energy-momentum tensor and baryon current from SMASH transport-model output. The workflow parallelizes SMASH event generation, constructs global regular or adaptive grids, groups particles into representative particles using box or k-means clustering, and then calculates tensors and currents via Gaussian smearing or coarse-graining. The central claim is that dynamic box clustering combined with smearing reproduces the no-cluster smearing baseline to 'perfect agreement' while significantly reducing computational cost, making the framework suitable for event-averaged thermodynamic properties and, potentially, hydrodynamic initialization or thermal emission calculations at high baryon density.","tokens_in":16838,"tokens_out":6903,"duration_ms":74188,"significance":"The framework addresses a genuine computational bottleneck: converting millions of transport-model particles into continuous fields for thousands of events. Its strengths include a clean modular workflow, a hash-map-based dynamic box clustering algorithm with O(N) scaling, and explicit parallelization. If the accuracy claims were fully established, the code would be a useful community tool for event-averaged backgrounds at BES energies. However, the reported validation tests only a restricted set of observables and does not establish preservation of the full energy-momentum tensor, which is required for the hydrodynamics and photon-emission applications advertised. The central claim is therefore currently too strong.","major_comments":[{"comment":"The representative-particle construction defined by Eqs. (1)-(2) does not preserve the spatial stress components of Eq. (3). Since the tensor is quadratic in momentum, with T^{ij} = sum_i p_i^j p_i^k / p_i^0, summing four-momenta first and then evaluating the tensor drops the intra-cluster momentum variance. For example, two identical particles at the same space-time point with momenta (+p,0,0) and (-p,0,0) give T^{xx} = 2p^2/E in the original cluster, while the representative particle with P = (2E,0,0,0) gives T^{xx} = 0. In a locally thermalized system this missing variance is precisely the pressure and shear stress. The validations in Sec. 4.1 and Fig. 3 use weighted averages of energy density, baryon density, transverse flow, and eccentricities at z=0, which are dominated by T^{00}, T^{0i}, and energy-density geometry; they do not test T^{ij}. Thus the Sec. 4.3 statement that 'Box cluster (dyn) + Smear' achieves 'perfect agreement' with the baseline is not established for the full energy-momentum tensor, and the method cannot yet be advertised as safe for hydrodynamic initialization or pressure-gradient-sensitive observables.","section":"Sec. 3.2.3, 3.3, and 4.3"},{"comment":"The quantitative validation is too thin to support the paper's accuracy claims. The comparison covers a single collision energy and an unspecified impact-parameter range; Sec. 4.1.1 explicitly omits numerical parameters ('we omit exhaustive details here'). Fig. 3 shows only weighted averages at z=0, with no standard deviations (their omission is acknowledged in Sec. 4.1.2) and no numerical agreement metric; phrases such as 'visually indistinguishable' and 'perfect agreement' are not quantified. The paper should report, at minimum, a component-wise comparison of T^{mu nu} and J^mu over the full three-dimensional grid between the baseline and each clustered case, including off-diagonal spatial components, together with a sensitivity study over the box size, smearing width, grid spacing, and k. Without these, the central claim that clustering preserves physical accuracy is not supported.","section":"Sec. 4.1 and Fig. 3"},{"comment":"Some reported complexities are algorithm-choice dependent and are not benchmarked. The O(N_particles x M_grid) cost quoted for 'No cluster + Coarse' reflects checking every particle against every grid point; a standard cell-list binning approach is O(N_particles + M_grid). The timing comparison in Fig. 4 includes only three of the six evaluated cases, so it does not substantiate the efficiency ordering among all clustering variants. Please report separate timings for clustering and smearing, and include the two box-clustering variants, so the efficiency claims are matched to the accuracy claims.","section":"Sec. 4.2.2 and Fig. 4"}],"minor_comments":[{"comment":"The text says k-means clusters particles based on 'spatial and momentum distributions', but Algorithm 2 takes only 3D spatial points and energies; it is unclear whether momentum coordinates enter the distance metric. Please clarify and discuss the dependence on the random initialization of centroids.","section":"Sec. 3.2.2 and Algorithm 2"},{"comment":"The energy-weighted coordinate definition in Eq. (2) is frame dependent; please state explicitly that representative coordinates are defined in the computational frame, and that only the four-momentum sum in Eq. (1) is Lorentz covariant.","section":"Sec. 3.2.3, Eq. (2)"},{"comment":"Please include a table with all numerical settings: collision energy, impact parameter selection, number of events, grid spacing and size, smearing width, box sizes, k, max_iters, and thread count. This is essential for reproducibility in a Computer Physics Communications paper.","section":"Sec. 4.1.1"},{"comment":"The phrase 'perfect agreement' in Sec. 4.3 is inconsistent with 'almost identical' in Sec. 4.1.4; please use a single quantified wording throughout.","section":"Sec. 4.3 and Sec. 4.1.4"},{"comment":"The statement that the unsmeared cases have 'much larger standard deviations which are not shown' makes it impossible to judge whether differences between cases are significant; please show the standard deviations or provide an error analysis.","section":"Sec. 4.1.2"}],"recommendation":"major_revision","confidential_remarks":"The manuscript fits the journal's scope and I see no citation or novelty concerns. The main technical obstacle is the unvalidated preservation of the spatial stress tensor; the stress-test concern about Eq. (3) lands, because the displayed observables in Fig. 3 do not constrain T^{ij}. I recommend major revision rather than rejection: the authors should either add a component-wise validation of the full tensor and temper the claims, or explicitly restrict the framework's advertised domain to energy density, baryon density, and flow fields. The current 'perfect agreement' wording will mislead users who initialize hydrodynamics with these tensors."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a genuinely useful piece of computational tooling, and I'd like to see it in CPC after revision. But the paper overstates what the validation shows. The two-particle example in the stress-test is exactly right: with momenta +p and -p, the original cluster has T^xx = 2 p_x^2/E, the representative has zero. Eq. (3) is quadratic in momentum, so replacing a cluster by its summed four-momentum throws away exactly the momentum-space variance that becomes pressure and shear stress. The paper admits the nonlinearity in Sec. 4.1.3 but then validates the representative-particle idea only through transverse-plane weighted averages of e, n_B, u_r, epsilon_2, epsilon_3 at z=0. Those are dominated by T^00, T^0i, and the energy-density geometry; they do not constrain T^ij. So \"perfect agreement with the baseline\" in Sec. 4.3 is not established for the full tensor, and the method is not ready for hydrodynamic initialization or any observable that depends on pressure gradients.\n\nWhere the paper earns its keep: the dynamical box cluster with hash maps is a real algorithmic improvement over iterative k-means and regular-box clustering; O(N_particles) clustering plus K_rep << N is the right way to attack the bottleneck. The modular workflow (parallel SMASH, global grid, parallel TensorCalc) is sensible, and the benchmark numbers show substantial speedups. The comparison of smearing vs coarse-graining is useful, and the discussion of why smearing increases epsilon_2 and reduces epsilon_3 is a nice physical touch.\n\nSoft spots beyond T^ij: the validation is one collision energy, one centrality class, z=0 only; error bars are explicitly not shown; \"perfect agreement\" is visual, not quantitative; no code release; the adaptive grid described in Sec. 3.1.2 is never tested (Sec. 4 uses a uniform grid). These are fixable. If the author releases the code and adds a quantitative comparison of T^xx, T^xy, etc. for at least one or two energies, the core claim becomes defensible. If not, the abstract and Sec. 4.3 need to be rewritten to claim accuracy only for the energy/baryon sector.\n\nBottom line: worth a serious referee. I'd accept it for review with the expectation of major revision. The engineering is solid; the physics claims need to be matched to what was actually computed.","headline":"Useful engineering for a real bottleneck, but the headline claim that representative particles preserve the energy-momentum tensor is only shown for the energy/baryon sector, not for spatial stress.","tokens_in":17350,"tokens_out":2153,"would_cite":true,"duration_ms":23372,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["25.75.-q","25.75.Nq"],"model":"deepseek-v4-flash","headline":"Clustered super-particles reproduce heavy-ion flow fields at a fraction of the cost.","keywords":["heavy-ion collisions","QCD matter","high baryon density","transport models","energy-momentum tensor","particle clustering","representative particles","smearing kernels"],"falsifier":"Compare the full local tensor $T^{\\mu\\nu}(t, \\mathbf{r})$ and its spatial gradients from \"No cluster + Smear\" and \"Box cluster (dyn) + Smear\" on identical SMASH events, looking for point-wise differences larger than the statistical uncertainty. A concrete version: compare the off-midplane components and the gradient $\\nabla T^{00}$ at grid cells away from $z = 0$, or compute thermal photon and dilepton spectra from the two tensors; if the spectra differ by more than a few percent, the representative-particle equivalence breaks down precisely where gradients feed emission.","tokens_in":16331,"feed_emoji":"⚛️","tokens_out":7434,"duration_ms":67579,"temperature":0.7,"pith_summary":"The paper introduces MATRICS, a modular workflow that turns the discrete particle lists produced by transport simulations of heavy-ion collisions into continuous space-time fields of energy, momentum, and baryon number. Its central claim is that replacing a cluster of nearby particles with a single representative particle that carries the cluster's summed four-momentum and an energy-weighted average position reproduces the smeared energy-momentum tensor and baryon current, provided the particles are grouped into space-time boxes. In particular, the \"Box cluster (dyn) + Smear\" variant is reported to agree with the baseline no-clustering smeared calculation while dropping the smearing cost from scaling with the number of particles to scaling with the much smaller number of clusters. If this holds, heavy-ion simulations at high baryon density can compute thermodynamic backgrounds for photon or dilepton emission and hydrodynamic initial conditions with far less computing time and memory.","feed_headline":"Super-particles cut heavy-ion tensor cost without losing accuracy","feed_subtitle":"A dynamic box-clustering method claims to match full smearing while shrinking the calculation by orders of magnitude in particle-grid pairs.","key_machinery":"The load-bearing object is the representative particle: for each space-time cluster, one particle whose four-momentum is the sum of the cluster's four-momenta, $p_{\\rm rep}^{\\mu} = \\sum_i p_i^{\\mu}$, and whose position is the energy-weighted space-time average, $x_{\\rm rep}^{\\mu} = \\sum_i p_i^0 x_i^{\\mu} / \\sum_i p_i^0$. This object carries the argument because it reduces the smearing sum from $O(N_{\\rm particles} \\times M_{\\rm grid})$ to $O(K_{\\rm rep} \\times M_{\\rm grid})$, with $K_{\\rm rep} \\ll N_{\\rm particles}$, and because the paper claims the continuous fields built from these super-particles coincide with those built from the original particles. The supporting machinery consists of a hash-map-based dynamical box clustering that scales as $O(N_{\\rm particles})$, the smearing kernel $K(t, \\mathbf{r}; t_i, \\mathbf{r}_i)$ in Eq. (3) that turns discrete particle contributions into continuous tensors, and the Landau matching procedure that converts $T^{\\mu\\nu}$ and $J^{\\mu}$ into energy density, baryon density, and flow velocity.","core_discovery":"The paper's central claim is that a cluster of $N$ particles can be replaced by a single representative particle with four-momentum $p_{\\rm rep}^{\\mu} = \\sum_i p_i^{\\mu}$ and space-time position $x_{\\rm rep}^{\\mu} = \\sum_i p_i^0 x_i^{\\mu}/\\sum_i p_i^0$, and that this replacement, when combined with a smearing kernel, preserves the physics of the original particle ensemble. Because the tensor formula $T^{\\mu\\nu} = p^{\\mu}p^{\\nu}/p^0$ is quadratic in momentum, summing tensors of individual particles differs in principle from evaluating the tensor of the summed momenta; the paper argues that this nonlinearity is inconsequential for the averaged thermodynamic quantities and geometric anisotropies it examines. The quantitative evidence comes from SMASH events at a fixed collision energy and impact-parameter range, comparing six combinations of clustering and tensor-calculation methods against the \"No cluster + Smear\" baseline for energy density, baryon density, transverse flow, eccentricity, and triangularity at $z = 0$. The headline result is that box clustering on a dynamically generated grid followed by smearing (\"Box cluster (dyn) + Smear\") is reported to agree with the baseline while being substantially cheaper than the unclustered calculation.","pith_inferences":["The paper's validation is limited to transverse-plane averages at $z = 0$; whether representative particles also preserve the local tensor structure and gradients that feed photon emission and gradient-driven observables remains untested, and a full 3D comparison would settle it.","The same clustering logic could be applied across events, collapsing super-particles from different events into larger clusters; this would trade event-by-event fluctuation information for additional speed.","If the representative-particle equivalence holds locally, the method should transfer directly to other conserved currents and derived fields, such as strangeness current, electromagnetic fields, or vorticity, with the same box-clustering procedure.","A natural stress test is to push the box size up until deviations from the baseline appear; the point where the approximation fails would define a resolution criterion for future users."],"forward_implications":["The \"Box cluster (dyn) + Smear\" method claims to reproduce the baseline smeared tensor while dropping the smearing cost from $O(N_{\\rm particles} \\times M_{\\rm grid})$ to $O(K_{\\rm rep} \\times M_{\\rm grid})$, a large speedup for events with millions of particles.","The five averaged observables tested at $z = 0$ — energy density, baryon density, transverse flow, eccentricity, and triangularity — are reported to match the no-clustering baseline, so event-averaged thermodynamic backgrounds can be built from clustered particles.","Because the workflow supports both regular grids (needed for hydrodynamic initial conditions) and adaptive grids (for thermal emission and isotropization studies), the same clustered-particle machinery serves both standard applications.","Memory use drops as well, since the dynamical box clustering stores only non-empty grid cells through hash maps.","The paper's proposed follow-up — an adaptive smearing grid built from representative-particle density — would further reduce the grid count in sparse regions and lower the smearing cost even more."],"supporting_citations":[{"why":"Defines the baseline smeared energy-momentum tensor from transport particles that the paper's no-cluster plus smear case follows.","marker":"[23]"},{"why":"Provides the smearing-kernel approach used for continuous tensor construction at low beam energies.","marker":"[24]"},{"why":"The SMASH transport model produces the discrete particle lists that all evaluation cases start from.","marker":"[25]"},{"why":"Supplies the coarse-grained transport method used in the no-cluster plus coarse comparison case.","marker":"[28]"},{"why":"Exemplifies the macroscopic thermal-emission use case where the clustered tensor serves as the thermodynamic background.","marker":"[30]"},{"why":"Hash-map data structure that gives the dynamical box clustering its O(N_particles) scaling.","marker":"[36]"},{"why":"k-means algorithm that underlies the energy-weighted dynamical clustering variant.","marker":"[37]"},{"why":"Justifies the energy-weighted average position used to define representative particle coordinates.","marker":"[38]"}],"fun_headline_variants":["Super-particles slash heavy-ion tensor cost, keep physics","Clustered particles speed up QCD matter calculations","MATRICS: faster heavy-ion thermodynamics via super-particles","Box clustering cuts tensor cost in baryon-rich QCD","Dynamic clustering: heavy-ion tensors without the compute"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument assumes that a single representative particle with summed four-momentum and energy-weighted coordinates reproduces the smeared energy-momentum tensor and baryon current at every space-time grid point, but the paper validates this only on weighted averages over the transverse plane at $z = 0$; because Eq. (3) is not linear in momentum, the equivalence is approximate rather than exact.","fun_headline_variants_meta":{"raw":{"variants":["Super-particles slash heavy-ion tensor cost, keep physics","Clustered particles speed up QCD matter calculations","MATRICS: faster heavy-ion thermodynamics via super-particles","Box clustering cuts tensor cost in baryon-rich QCD","Dynamic clustering: heavy-ion tensors without the compute"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000843,"raw_usage":{"total_tokens":3678,"prompt_tokens":958,"completion_tokens":2720,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":574,"completion_tokens_details":{"reasoning_tokens":2640}},"tokens_in":574,"tokens_out":2720,"duration_ms":19923,"temperature":1.0,"reasoning_tokens":2640,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T18:24:23.357654+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compare the full local tensor $T^{\\mu\\nu}(t, \\mathbf{r})$ and its spatial gradients from \"No cluster + Smear\" and \"Box cluster (dyn) + Smear\" on identical SMASH events, looking for point-wise differences larger than the statistical uncertainty. A concrete version: compare the off-midplane components and the gradient $\\nabla T^{00}$ at grid cells away from $z = 0$, or compute thermal photon and dilepton spectra from the two tensors; if the spectra differ by more than a few percent, the representative-particle equivalence breaks down precisely where gradients feed emission.","supporting_citations":[{"cited_title":"Dilepton production and reaction dynamics in heavy-ion collisions at SIS energies from coarse-grained transport simulations","cited_arxiv_id":"1505.06131","evidence_quote":"Exemplifies the macroscopic thermal-emission use case where the clustered tensor serves as the thermodynamic background."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Hash-map data structure that gives the dynamical box clustering its O(N_particles) scaling."},{"cited_title":"Likas, N","cited_arxiv_id":null,"evidence_quote":"k-means algorithm that underlies the energy-weighted dynamical clustering variant."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Justifies the energy-weighted average position used to define representative particle coordinates."}],"review_version":2}