{"id":"924abe38-d05c-4121-b477-82b62948b2d2","arxiv_id":"2411.19470","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A deep neural network extends the DRHBc nuclear mass table to odd-Z nuclei, and r-process simulations show that the resulting mass differences, attributed to deformation, strongly affect abundances around A=80-120.","lead":"The authors train a neural network to predict nuclear masses for odd-proton nuclei using the DRHBc model, then use the extended mass table to test whether nuclear deformation shapes r-process element abundances. They report that r-process yields differ by up to two orders of magnitude in the mass range A=80-120, particularly for magnetohydrodynamic jet environments.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Deformation attribution is confounded: RCHB* and DRHBc* differ in pairing strength (-342.5 vs -325.0 MeV fm^3), so the A=80-120 abundance shifts cannot be assigned to deformation without a controlled same-pairing comparison.","rationale":"The reader's weakest assumption correctly identifies the deformation-versus-pairing confound, and this is the most load-bearing issue in the paper. The paper explicitly states different pairing strengths for RCHB and DRHBc (Section I), yet Section V asserts that deformation is the primary difference without a control calculation. Since the r-process sensitivity study compares only these two tables, any abundance difference could be due to the pairing-strength mismatch. The DNN extension has some independent support from the out-of-sample comparison with 34 newly measured isotopes (RMS 0.725 MeV), so the mass-extension part is not the primary fatal flaw. However, the central astrophysical conclusion about deformation sensitivity is unsupported in its current form. The proposed test—varying pairing strength while holding deformation and other inputs fixed—would directly settle whether the abundance differences survive after removing the confound. Because this test is not present in the manuscript, the rejection verdict remains appropriate, and no change to the reader's verdict is needed.","tokens_in":12146,"tokens_out":4479,"duration_ms":40439,"concrete_test":"Recompute the DRHBc mass table for the nuclei in Table III (or the full >5 MeV set) with the pairing strength changed from -325.0 to -342.5 MeV fm^3, keeping all other inputs identical, and rerun the MHD and collapsar abundance calculations with this rescaled DRHBc table against RCHB*. If the A=80-120 abundance differences largely disappear, the deformation attribution fails; if they persist, the confound is resolved. A complementary spherical-DRHBc run with the same pairing strength would further isolate the deformation effect directly.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central astrophysical claim—that r-process abundances are sensitive to nuclear deformation—rests on comparing the RCHB* and DRHBc* mass tables. However, these tables differ by more than deformation: Section I lists pairing strengths of -342.5 MeV fm^3 for RCHB and -325.0 MeV fm^3 for DRHBc. Section V then asserts that 'the primary difference between the two mass tables lies in nuclear deformations' without presenting any control calculation that holds pairing fixed. Because pairing strength directly changes binding energies, beta-decay Q-values, and neutron-capture Q-values, the abundance differences in Figs. 3-5 could be driven by the pairing-strength mismatch rather than by deformation. The post-hoc selection of nuclei with >5 MeV mass differences and the DNN interpolation using AME2020 odd-Z data do not remove this confound. The out-of-sample RMS of 0.725 MeV against newly measured masses (Table II) is a genuine positive result, but it validates total mass predictions, not the attribution of the RCHB* versus DRHBc* differences to deformation. As written, the paper's headline conclusion is not supported by a controlled comparison.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper uses a deep neural network (DNN) to extend the deformed relativistic Hartree-Bogoliubov in continuum (DRHBc) mass table, which is otherwise available only for even-Z nuclei, by training on DRHBc even-Z binding energies together with AME2020 experimental masses for odd-Z and odd-odd nuclei. The resulting DRHBc* and RCHB* tables are then used in r-process network calculations for MHD jet and collapsar jet environments, selecting nuclei where the two mass tables differ by more than 5 MeV. The authors report an RMS deviation of 0.842 MeV between DRHBc* and AME2020 and find r-process abundance differences of up to two orders of magnitude in the A=80-120 region, which they attribute to nuclear deformation.","tokens_in":12376,"tokens_out":6168,"duration_ms":53921,"significance":"If the deformation attribution were cleanly established, this would be a useful contribution to the discussion of deformation effects in r-process nucleosynthesis, and the DNN extension of the DRHBc table addresses a genuine gap. The out-of-sample comparison against newly measured isotopes in Table II is a genuine positive result: those 34 nuclei were not in AME2020, and the reported 0.725 MeV RMS is informative, albeit for a narrow region (Z=32-37). The manuscript also makes good use of two astrophysical sites, MHD and collapsar jets, and includes mass-only, mass-plus-beta, mass-plus-neutron-capture, and combined sensitivity cases. However, the significance is substantially weakened by the circularity of the main DNN performance metric and by the lack of a controlled comparison between the spherical and deformed mass tables.","major_comments":[{"comment":"The reported RMS deviation of 0.842 MeV for DRHBc* is an in-sample training error, not a predictive test. Section II states that the DNN uses AME2020 binding energies as part of the training set, and Section III then compares the predicted masses with the same AME2020 data. The comparison with the even-Z DRHBc value of 1.433 MeV is therefore misleading, because that value is an out-of-sample deviation of the original DRHBc table, whereas 0.842 MeV measures how well the network reproduces data it was trained on. The only out-of-sample evidence is Table II, with 34 newly measured isotopes and an RMS of 0.725 MeV; this should be presented as the primary DNN validation, and the in-sample RMS should not be quoted as an improved predictive accuracy.","section":"Section III, Table I"},{"comment":"The central attribution of the r-process abundance differences to nuclear deformation is confounded by the different pairing strengths of the two base models. Section I explicitly states that the pairing strength is -342.5 MeV fm^3 in RCHB and -325.0 MeV fm^3 in DRHBc. Section V asserts that 'the primary difference between the two mass tables lies in nuclear deformations' without providing any control calculation that holds the pairing strength fixed. Since pairing strength directly affects binding energies, beta-decay Q-values, and neutron-capture Q-values, the abundance differences seen in Figs. 3-5 could be driven entirely or partly by the pairing mismatch rather than by deformation. The manuscript needs a same-pairing spherical-versus-deformed comparison, or an equivalent explicit control, before the deformation conclusion can be drawn.","section":"Section V and Section I"},{"comment":"The sensitivity study uses mass differences between RCHB* and DRHBc* that exceed 5 MeV, but both starred tables are DNN products trained partly on AME2020. For the selected exotic nuclei, the mass differences in Figure 2 and Table III are therefore DNN extrapolations (or interpolations), not direct outputs of either RCHB or DRHBc. This means the selected differences may include DNN prediction error rather than a physical deformation effect. The authors note that the 5 MeV threshold is larger than the 1-2 MeV variations used in prior sensitivity studies (Refs. [8,10]), but no uncertainty quantification is provided for the DNN-extrapolated masses, so the significance of the two-order-of-magnitude abundance changes cannot be assessed reliably.","section":"Section IV and Figure 2"},{"comment":"The DNN architecture and training details are not reported. Equations (1)-(5) define a generic feed-forward network with an unspecified number of hidden layers and nodes, and no values are given for the number of layers, number of nodes per layer, learning rate, number of epochs, batch size, regularization, or train/validation split. The statement that 'we use a deep neural network consisting of the first hidden layer, intermediate hidden layer, last hidden layer and output layer' is not sufficient to reproduce the results. Given that the DNN extension is one of the two central contributions, these details are necessary for reproducibility.","section":"Section II"},{"comment":"There is an internal inconsistency in the reported RCHB* RMS values. The text first gives RMS deviations of 1.599 MeV and 2.457 MeV for the comparison with AME2020 and with RCHB+AME2020, respectively, but later states that with the five-input case the corresponding RMS deviations are 1.862 MeV and 1.957 MeV. Table I lists yet another set of values (1.779, 1.584, and 1.862 MeV for two-, four-, and five-input RCHB*). The reader cannot determine which numbers correspond to which input configuration, and this discrepancy should be resolved.","section":"Section II (validation procedure)"}],"minor_comments":[{"comment":"The caption says the quadrupole deformation beta2 values are 'taken from AME2020 [39] and from FRDM(2012) [56]', but AME2020 is a mass evaluation and does not provide deformation parameters. The source of the first beta2 column should be clarified.","section":"Table III caption"},{"comment":"The sentence 'For comparison, the RMS deviation between AME2020 and the even-Z DRHBc⋆ mass table is 1.433 MeV' conflates the original even-Z DRHBc mass table with the DNN-produced DRHBc* table. Since the DNN output is not the same as the original DRHBc table, the terminology should distinguish them clearly.","section":"Section III"},{"comment":"The text has a typo: 'Eref i denotes the binding available binding energies' should be 'Eref i denotes the available binding energies'.","section":"Equation (7)"},{"comment":"The conclusion states that r-process abundances are sensitive to 'nuclear deformation', but the study actually compares two mass tables that differ in pairing strength, deformation, and DNN extrapolation. The wording should be softened to 'mass sensitivity' unless the deformation control is added.","section":"Abstract and Section V"}],"recommendation":"major_revision","confidential_remarks":"The reader's reject verdict is understandable, and the pairing-strength confound is a serious flaw in the astrophysical conclusion. I chose major_revision rather than reject because the confound could in principle be addressed by a controlled same-pairing spherical-versus-deformed calculation, and because the Table II out-of-sample DNN result is a real positive that would be worth publishing with a properly non-circular performance metric. The editor may wish to require the authors to either add the control calculation and a genuine train/test split, or restrict the conclusions to what the current evidence supports."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The real substance here is the DNN extension of the DRHBc mass table to odd-Z nuclei, and the out-of-sample validation is the strongest part of the paper. Comparing against 34 newly measured neutron-rich masses, including first measurements of 88,89As, and getting 0.725 MeV RMS is a legitimate positive result. It tells me the DNN-plus-AME2020 interpolation is doing something sensible for exotic nuclei, not just memorizing the training set. The Sn plots also show the expected odd-even staggering, which adds credibility.\n\nThe soft spots are real, though. Table I's 0.842 MeV RMS is not a predictive test because AME2020 odd-Z masses are in the training set. That is a training-set comparison, and the paper should say so more prominently. The out-of-sample check partially rescues it, but it is a small sample and does not cover the full Z,N range of the claimed table.\n\nThe bigger issue is the deformation attribution. RCHB and DRHBc differ in pairing strength too: -342.5 vs -325.0 MeV fm^3. Section V says the primary difference lies in nuclear deformations, but there is no controlled comparison holding pairing fixed. So the abundance shifts in Figs. 3-5 could come from the pairing mismatch, or from a combination. The paper actually hedges in places—calling it a sensitivity to \"deformations (or masses)\"—but the abstract and summary land on deformation without the necessary caveat. That needs a fix, either a same-pairing control calculation or a tempered conclusion.\n\nThe 5 MeV mass-difference threshold is also large and post-hoc. The authors acknowledge this compared to standard ±0.5-1 MeV sensitivity studies, but it means the result is a big-stick probe, not a realistic uncertainty quantification. That is okay for a preliminary study, but worth stating more clearly.\n\nOne thing I like: the collapsar result shows fission recycling almost eliminates the differences, which is a useful nuance and suggests the sensitivity is environment-dependent.\n\nWho is this for? Nuclear structure people working on mass tables and r-process modelers who want a look at how different mass models propagate. It is a preliminary study, but it identifies a concrete path: finish the full DRHBc table and do this properly.\n\nMy take: the paper deserves peer review, but as major revision. The out-of-sample validation earns referee time, and the central claim just needs a controlled comparison or a more honest scope statement.","headline":"A useful DNN-completed odd-Z DRHBc mass table with a genuine out-of-sample check, but the r-process sensitivity claim is a mass-difference study between models differing in pairing as well as deformation, so the deformation attribution needs a control.","tokens_in":12976,"tokens_out":2728,"would_cite":false,"duration_ms":26010,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The shape of a nucleus can change r-process element yields by a factor of 100.","keywords":["deep neural network","nuclear mass table","DRHBc","RCHB","r-process nucleosynthesis","nuclear deformation","mass sensitivity","neutron-rich nuclei"],"falsifier":"Compute the same r-process abundance comparison using RCHB and DRHBc versions with identical pairing strength; if the A = 80-120 abundance differences disappear, the observed sensitivity is due to pairing, not deformation. Alternatively, measure the masses of the key deformed isotopes such as 93As with high precision and check which mass table, RCHB* or DRHBc*, agrees with experiment.","tokens_in":11906,"feed_emoji":"⚛️","tokens_out":2990,"duration_ms":25023,"temperature":0.7,"pith_summary":"This paper asks whether nuclear deformation, not just nuclear mass, matters for the astrophysical r-process that builds half of the heavy elements. To answer it, the authors first use a deep neural network to extend the deformed DRHBc mass table to odd-Z nuclei, reaching an RMS deviation of 0.842 MeV against AME2020. They then feed two mass tables, one spherical (RCHB*) and one deformed (DRHBc*), into r-process network calculations for magnetohydrodynamic jets and collapsar jets. The resulting abundance patterns differ by up to two orders of magnitude in the mass range A = 80-120, leading the authors to conclude that deformation is a significant input for r-process predictions.","feed_headline":"Deformed nuclei move r-process yields by two orders of magnitude","feed_subtitle":"A neural network completes the DRHBc mass table, letting models trace abundance shifts to nuclear shape.","key_machinery":"The load-bearing tool is a deep neural network with five inputs (Z, N, pairing term delta, and shell-effect terms Pn and Pp) that interpolates binding energies for odd-odd and odd-even nuclei using even-Z DRHBc results plus AME2020 data. The sensitivity analysis then rests on the mass difference between the spherical RCHB* and deformed DRHBc* tables for a set of neutron-rich isotopes in the A = 80-120 range, where the difference exceeds 5 MeV. These mass shifts change neutron-capture Q-values, which in turn alter reaction rates and shift the abundance pattern; for example, the Q-value of 90As(n,gamma)91As changes from 1.04 MeV to 6.22 MeV between the two tables.","core_discovery":"The central claim is that r-process abundances are sensitive to nuclear deformation, particularly in the mass region A = 80-120. This is demonstrated by comparing abundance yields computed with the spherical RCHB* mass table and the axially deformed DRHBc* mass table in two r-process sites, magnetohydrodynamic jets and collapsar jets. In the MHD jet model the yields differ by up to two orders of magnitude in that mass window, with the deformed table producing more neutron-rich nuclei beyond A = 92. A supporting result is that a five-input deep neural network, using proton number, neutron number, pairing, and separate proton and neutron shell effects, can extend the even-Z DRHBc table to odd-Z nuclei with an RMS deviation of 0.842 MeV from AME2020.","pith_inferences":["If the deformation sensitivity is confirmed, r-process abundance patterns could become a probe of the ground-state deformations of exotic neutron-rich nuclei in the A = 80-120 region.","The DNN extension is only as reliable as its training data; future mass measurements of neutron-rich odd-Z nuclei such as As, Ga, and Se isotopes would test whether the 0.842 MeV RMS deviation holds outside the training set.","The pairing strength is different in the two base models (DRHBc uses -325.0 MeV fm^3, RCHB uses -342.5 MeV fm^3), so a cleaner test of the deformation claim would compare two tables built with identical pairing parameters.","The same DNN approach could be applied to separately predict beta-decay Q-values or neutron-capture rates, not just masses, to propagate deformation effects more directly into the reaction network."],"forward_implications":["A complete DRHBc mass table, extended by machine learning, becomes usable for r-process nucleosynthesis studies that include odd-Z nuclei.","Nuclear deformation should be treated as a first-order input in r-process sensitivity analyses, not just an adjustment to mass.","The abundance patterns from MHD jet models differ by up to two orders of magnitude between spherical and deformed mass tables, so deformation can change predicted yields of elements near A = 80-120.","In collapsar jet environments, fission recycling largely erases the deformation-induced differences at A = 100-120, showing that the astrophysical site determines how strongly deformation is imprinted on final abundances.","Improved solar r-abundance measurements near A = 104 and a complete DRHBc table would allow a direct test of the deformation-abundance connection."],"supporting_citations":[{"why":"Provides the even-Z DRHBc mass table that the DNN extends to odd-Z nuclei.","marker":"[38]"},{"why":"Supplies AME2020 binding energies used as training data and as the reference for RMS deviation.","marker":"[39]"},{"why":"Supplies FRDM(2012) quadrupole deformations used to identify the selected isotopes as deformed.","marker":"[56]"},{"why":"Provides the MHD jet trajectories and fission rates used in the r-process network calculation.","marker":"[70]"},{"why":"Provides the collapsar jet trajectories used for the second r-process site.","marker":"[72]"},{"why":"Lists newly measured neutron-rich isotope masses, including first measurements of 88As and 89As, used to validate DRHBc* predictions.","marker":"[55]"},{"why":"Establishes the baseline sensitivity of r-process abundances to mass variations of about 1 MeV, which this paper's 5 MeV differences exceed.","marker":"[8]"}],"fun_headline_variants":["R-process yields shift by 100x with deformed nuclear masses","Deformed masses alter r-process abundances in A=80-120","Neural-network mass table links nuclear shape to r-process yields","Spherical vs deformed: r-process abundances diverge by 100x"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The paper assumes that the mass difference between the RCHB* and DRHBc* tables is caused primarily by nuclear deformation, even though the two base models also use different pairing strengths (-342.5 vs -325.0 MeV $fm^{3}$).","fun_headline_variants_meta":{"raw":{"variants":["R-process yields shift by 100x with deformed nuclear masses","Deformed masses alter r-process abundances in A=80-120","Neural-network mass table links nuclear shape to r-process yields","Spherical vs deformed: r-process abundances diverge by 100x"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000402,"raw_usage":{"total_tokens":2122,"prompt_tokens":997,"completion_tokens":1125,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":613,"completion_tokens_details":{"reasoning_tokens":1051}},"tokens_in":613,"tokens_out":1125,"duration_ms":8443,"temperature":1.0,"reasoning_tokens":1051,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T10:09:12.475856+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the same r-process abundance comparison using RCHB and DRHBc versions with identical pairing strength; if the A = 80-120 abundance differences disappear, the observed sensitivity is due to pairing, not deformation. Alternatively, measure the masses of the key deformed isotopes such as 93As with high precision and check which mass table, RCHB* or DRHBc*, agrees with experiment.","supporting_citations":[{"cited_title":"Guo et al","cited_arxiv_id":null,"evidence_quote":"Provides the even-Z DRHBc mass table that the DNN extends to odd-Z nuclei."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies AME2020 binding energies used as training data and as the reference for RMS deviation."},{"cited_title":"M¨ oller, A","cited_arxiv_id":null,"evidence_quote":"Supplies FRDM(2012) quadrupole deformations used to identify the selected isotopes as deformed."},{"cited_title":"Shibagaki, T","cited_arxiv_id":null,"evidence_quote":"Provides the MHD jet trajectories and fission rates used in the r-process network calculation."},{"cited_title":"Nakamura, T","cited_arxiv_id":null,"evidence_quote":"Provides the collapsar jet trajectories used for the second r-process site."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Lists newly measured neutron-rich isotope masses, including first measurements of 88As and 89As, used to validate DRHBc* predictions."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the baseline sensitivity of r-process abundances to mass variations of about 1 MeV, which this paper's 5 MeV differences exceed."}],"review_version":1}