Pith. sign in

REVIEW 3 major objections 2 minor 43 references

A Numerical PDEs Approach to Evolution Equations in Shape Analysis Based on Regularized Morphoelasticity

T0 review · 3 major / 2 minor · reviewed 2026-07-13 · grok-4.5

Challenge this review Download PDF Read on arXiv ↗

Pith's one-line read High-order regularization turns the continuous inverse morphoelastic growth problem into a high-order elliptic system that mixed finite elements can solve efficiently.

desk verdict Only the abstract matches the morphoelasticity title; the body is an unrelated LLM-unlearning paper, so we have a methods pitch with zero verifiable math or numerics. read the letter →

arxiv 2604.04984 v5 submitted 2026-04-05 math.NA cs.NAmath.OC

Ziqin Zhou This is my paper
classification math.NAcs.NAmath.OC MSC 65N3074B0549J2035J48
keywords morphoelasticityshapeevolutionhigh-orderellipticregularizationmixedfiniteelementsLDDMMgrowthtensoroptimalcontrolFEniCSx
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper addresses shape evolution under volumetric growth by extending linear elasticity so that total deformation splits into an elastic part and a growth tensor G. The forward map (growth to displacement) is classical, but the continuous inverse problem of recovering a growth path that takes one configuration to another is ill-posed. The author embeds parametrized growth models inside an LDDMM-inspired optimal-control problem that treats growth as a path in shape space and seeks the energy-minimizing trajectory. A high-order regularization term is added, lifting the governing equations to a high-order elliptic system that possesses smooth solutions. The dissertation then concentrates on making that system computationally tractable: mixed finite-element discretizations reduce it to a coupled second-order system that can be solved with standard FEniCSx solvers. The result is a practical numerical pipeline for regularized morphoelastic shape matching.

What carries the argument

The high-order regularization term (which produces a high-order elliptic operator) together with its mixed finite-element reformulation into a system of coupled second-order equations; these two devices restore well-posedness and enable practical solution of the inverse growth problem.

What would settle it

Show that either the mixed finite-element scheme fails to converge for the regularized system on a simple growth benchmark, or that the computed energy-minimizing growth path produces configurations whose elastic residual is larger than that of a known biologically plausible alternative.

Watch

Extended reading notes

Core claim

A carefully chosen high-order regularization elevates the morphoelastic equilibrium equations to a high-order elliptic system whose smooth solutions exist; the same system can be rewritten as a mixed second-order system that standard finite-element libraries solve efficiently, thereby rendering the continuous LDDMM-style inverse growth problem computationally feasible.

Load-bearing premise

That the particular high-order regularizer plus the chosen parametrized growth models are enough to make the continuous inverse problem well-posed and that the resulting energy-minimizing path is the biologically or geometrically most plausible trajectory.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 2 minor

Summary. The submission is presented as a dissertation-style study of a regularized morphoelasticity problem for continuous shape evolution. From the abstract, the total deformation is split into elastic and growth parts via a growth tensor G; the forward map (displacement given G) is treated as standard linear elasticity with growth, while the continuous inverse problem of recovering a growth path is cast as an LDDMM-inspired optimal-control problem on shape space. A high-order regularization is claimed to lift the system to a high-order elliptic problem with smooth solutions, to be discretized by mixed finite elements (coupled second-order system) in FEniCSx. The body of the supplied manuscript, however, is an entirely different paper (CURE: circuit-aware unlearning for LLM-based recommendation), with no morphoelasticity, no high-order elliptic analysis, no mixed FEM, and no FEniCSx experiments.

Significance. If the abstract’s program were carried through—well-posed high-order regularized morphoelastic evolution, a clear LDDMM-style metric on growth paths, and a proven, efficient mixed FEM—the work would be a useful bridge between continuum growth models and computational anatomy. That significance cannot be assessed from the supplied full text, which does not contain the claimed analysis or numerics. The only available content is the abstract’s program statement; no theorems, error estimates, or benchmarks are present to credit.

major comments (3)
  1. Title/abstract vs. full text: the manuscript body is the unrelated CURE paper on LLMRec unlearning (circuits, forget/retain losses, MovieLens/GoodReads experiments). None of the abstract’s objects—growth tensor G, high-order morphoelastic elliptic system, mixed FEM, FEniCSx, or LDDMM-style optimal control—appear in the body. The central claims of the submission therefore have no supporting derivation or evidence in the document under review.
  2. Abstract claim that high-order regularization “ensures the existence of a smooth solution”: no weak/strong formulation, function spaces, or existence/uniqueness theorem is given anywhere in the supplied text. Without that analysis, the well-posedness premise of the continuous inverse problem remains an assertion.
  3. Abstract claim that mixed FEM makes the high-order system computationally feasible: there is no mixed variational form, no choice of finite-element spaces, no stability/error estimate, and no numerical table or figure for the morphoelastic problem. Efficiency and feasibility cannot be verified.
minor comments (2)
  1. Even within the abstract alone, the order of the high-order regularizer, the precise parametrization of G, and the form of the LDDMM-inspired cost are left unspecified, which would need fixing if a correct full manuscript were resubmitted.
  2. The abstract refers to the work as a “dissertation” while the header presents it as a research paper; scope and contribution level should be aligned in any resubmission.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; derivation chain (where present) is self-contained methodological comparison, not reduction to inputs by construction.

full rationale

The supplied full manuscript body is the CURE unlearning paper (arXiv 2604.04982), not the morphoelasticity/LDDMM PDE paper whose abstract and arXiv ID appear in the header. Within the actual body, the only load-bearing derivation is Theorem 5.1 comparing one-step CURE parameter updates against ordinary weighted gradient descent. The proof proceeds by Taylor expansion of the joint loss, algebraic rearrangement into parts A and B, a cosine condition controlled by explicit gradient normalization, and a sign argument that follows directly from the paper's own definition of forget/retain circuits via greedy edge selection on the same losses. None of these steps redefine the target quantity in terms of itself, fit a free parameter and then re-label the fit as a prediction, or rest on an unverified self-citation uniqueness claim. Circuit extraction (activation intervention / patching + PPR) and the subsequent selective update rules are constructive algorithmic choices, not circular. Experimental claims are empirical comparisons against external baselines, not forced by construction. Consequently the circularity score is 0; the abstract of the mismatched morphoelasticity paper likewise contains only a standard variational-regularization program with no self-referential reduction visible.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

Abstract-only review of a numerical PDE dissertation. Load-bearing modeling choices are standard continuum mechanics plus regularization/control assumptions; free parameters of growth models and regularization strength are implied but not quantified. No invented physical entities; the 'circuit' content in the supplied full text belongs to a different paper and is ignored.

free parameters (3)
  • Growth tensor / growth path parameters (parametrized G)
    Abstract states parametrized growth models inside an optimal-control problem; specific parametrization and any fitted coefficients are not given.
  • High-order regularization strength / order
    A high-order regularization term is introduced to obtain a high-order elliptic system; coefficient and exact differential order are unspecified in the abstract.
  • Weights balancing energy-efficient trajectory vs data fidelity in the optimal-control / LDDMM-inspired metric
    Seeking the most plausible energy-efficient path implies trade-off weights that are not specified.
assumptions (5)
  • domain assumption Total deformation decomposes multiplicatively/additively into elastic and growth components via growth tensor G (morphoelastic framework).
    Stated as foundation; standard in morphoelasticity literature, not derived here.
  • domain assumption Linear elasticity governing equations remain an adequate base when extended by volumetric growth for the regimes considered.
    Abstract builds on linear elasticity; geometric nonlinearity and large-strain issues are not addressed in the abstract.
  • ad hoc to paper High-order regularization elevates the system to a high-order elliptic problem that ensures existence of a smooth solution.
    Existence claim is asserted for the constructed regularized system; proof not available in abstract.
  • domain assumption An LDDMM-inspired shape-space path metric identifies the energy-minimizing growth trajectory as the physically meaningful inverse solution.
    Modeling choice that turns the ill-posed inverse problem into an optimal-control problem.
  • standard math Mixed FEM decomposition into coupled second-order equations is a valid and efficient treatment of the high-order system.
    Standard numerical strategy for high-order elliptic PDEs; effectiveness for this problem is a claim to be verified numerically.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Numerical PDEs Approach to Evolution Equations in Shape Analysis Based on Regularized Morphoelasticity." pith.science (2026). https://pith.science/paper/2604.04984

@misc{pith2026260404984,
  author       = {Pith},
  title        = {Pith review of: A Numerical PDEs Approach to Evolution Equations in Shape Analysis Based on Regularized Morphoelasticity},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2604.04984}},
  note         = {Machine review of arXiv:2604.04984}
}
abstract

This work studies a variational formulation and numerical solution of a regularized morphoelasticity problem of shape evolution. The foundation of our analysis is based on the governing equations of linear elasticity, extended to account for volumetric growth. In the morphoelastic framework, the total deformation is decomposed into an elastic component and a growth component, represented by a growth tensor $G$. While the forward one-step problem -- computing displacement given a growth tensor -- is well-established, a more challenging and relevant question in biological modeling is the inverse problem in a continuous sense. While this problem is fundamentally ill-posed without additional constraints, we will explore parametrized growth models inscribed within an optimal control problem inspired by the Large Deformation Diffeomorphic Metric Mapping (LDDMM) framework. By treating the growth process as a path within a shape space, we can define a physically meaningful metric and seek the most plausible, energy-efficient trajectory between configurations. In the construction, a high-order regularization term is introduced. This elevates the governing equations to a high-order elliptic system, ensuring the existence of a smooth solution. This dissertation focuses on the issue of solving this equation efficiently, as this is a key requirement for the feasibility of the overall approach. This will be achieved with the help of finite element solvers, notably from the FEniCSx library in Python. Also, we implement a Mixed Finite Element Method, which decomposes the problem into a system of coupled second-order equations as a treatment of these high-order systems that have significant computational challenges.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

43 extracted references · 13 linked inside Pith

  1. [1]

    Lucas Bourtoule, Varun Chandrasekaran, Christopher A Choquette-Choo, Hen- grui Jia, Adelin Travers, Baiwu Zhang, David Lie, and Nicolas Papernot. 2021. Machine unlearning. In2021 IEEE symposium on security and privacy (SP). IEEE, 141–159

    2021

  2. [2]

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. 2020. Language models are few-shot learners.Advances in neural information processing systems33 (2020), 1877–1901

    2020

  3. [3]

    Steven Cao, Victor Sanh, and Alexander M Rush. 2021. Low-complexity probing via finding subnetworks.arXiv preprint arXiv:2104.03514(2021)

    arXiv 2021

  4. [4]

    Chong Chen, Fei Sun, Min Zhang, and Bolin Ding. 2022. Recommendation unlearning. InProceedings of the ACM web conference 2022. 2768–2777

    2022

  5. [5]

    Chaochao Chen, Jiaming Zhang, Yizhao Zhang, Li Zhang, Lingjuan Lyu, Yuyuan Li, Biao Gong, and Chenggang Yan. 2024. CURE4Rec: A benchmark for recom- mendation unlearning with deeper influence.Advances in Neural Information Processing Systems37 (2024), 99128–99144

    2024

  6. [6]

    Ziheng Chen, Jiali Cheng, Hadi Amiri, Kaushiki Nag, Lu Lin, Sijia Liu, Gabriele Tolomei, and Xiangguo Sun. 2025. FROG: Fair Removal on Graph. InProceedings of the 34th ACM International Conference on Information and Knowledge Manage- ment(Seoul, Republic of Korea)(CIKM ’25). Association for Computing Machin- ery, New York, NY, USA, 415–424. https://doi.or...

    doi:10.1145/3746252.3761341 2025

  7. [7]

    Jiali Cheng and Hadi Amiri. 2024. Mu-bench: A multitask multimodal benchmark for machine unlearning.arXiv preprint arXiv:2406.14796(2024)

    arXiv 2024

  8. [8]

    Jiali Cheng and Hadi Amiri. 2025. MultiDelete for Multimodal Machine Un- learning. InComputer Vision – ECCV 2024, Aleš Leonardis, Elisa Ricci, Stefan Roth, Olga Russakovsky, Torsten Sattler, and Gül Varol (Eds.). Springer Nature Switzerland, Cham, 165–184

    2025

Show all 43 references
  1. [9]

    Jiali Cheng and Hadi Amiri. 2025. Tool unlearning for tool-augmented llms. arXiv preprint arXiv:2502.01083(2025)

    2025 arXiv

  2. [10]

    Jiali Cheng, Ziheng Chen, Chirag Agarwal, and Hadi Amiri. 2026. Toward Understanding Unlearning Difficulty: A Mechanistic Perspective and Circuit- Guided Difficulty Metric.arXiv preprint arXiv:2601.09624(2026)

    2026

  3. [11]

    Jiali Cheng, George Dasoulas, Huan He, Chirag Agarwal, and Marinka Zitnik

  4. [12]

    InThe Eleventh International Conference on Learning Representations

    GNNDelete: A General Strategy for Unlearning in Graph Neural Networks. InThe Eleventh International Conference on Learning Representations. https: //openreview.net/forum?id=X9yCkmT5Qrl

  5. [13]

    Arthur Conmy, Augustine Mavor-Parker, Aengus Lynch, Stefan Heimersheim, and Adrià Garriga-Alonso. 2023. Towards automated circuit discovery for mech- anistic interpretability.Advances in Neural Information Processing Systems36 (2023), 16318–16352

    2023

  6. [14]

    Chongyu Fan, Jinghan Jia, Yihua Zhang, Anil Ramakrishna, Mingyi Hong, and Si- jia Liu. 2025. Towards llm unlearning resilient to relearning attacks: A sharpness- aware minimization perspective and beyond.arXiv preprint arXiv:2502.05374 (2025)

    2025 arXiv

  7. [15]

    Chongyu Fan, Jiancheng Liu, Yihua Zhang, Eric Wong, Dennis Wei, and Sijia Liu

  8. [16]

    Salun: Empowering machine unlearning via gradient-based weight saliency in both image classification and generation.arXiv preprint arXiv:2310.12508 (2023)

    2023 arXiv

  9. [17]

    Michael Hanna, Ollie Liu, and Alexandre Variengien. 2023. How does GPT- 2 compute greater-than?: Interpreting mathematical abilities in a pre-trained language model.Advances in Neural Information Processing Systems36 (2023), 76033–76060

    2023

  10. [18]

    Michael Hanna, Sandro Pezzelle, and Yonatan Belinkov. 2024. Have faith in faithfulness: Going beyond circuit overlap when finding model mechanisms. arXiv preprint arXiv:2403.17806(2024)

    2024 arXiv

  11. [19]

    Zhiyu Hu, Yang Zhang, Minghao Xiao, Wenjie Wang, Fuli Feng, and Xiang- nan He. 2025. Exact and efficient unlearning for large language model-based recommendation.IEEE Transactions on Knowledge and Data Engineering(2025)

    2025

  12. [20]

    Farnoush Rezaei Jafari, Oliver Eberle, Ashkan Khakzar, and Neel Nanda. 2025. RelP: Faithful and Efficient Circuit Discovery in Language Models via Relevance Patching.arXiv preprint arXiv:2508.21258(2025)

    2025

  13. [21]

    Meghdad Kurmanji, Peter Triantafillou, Jamie Hayes, and Eleni Triantafillou

  14. [22]

    Towards unbounded machine unlearning.Advances in neural information processing systems36 (2023), 1957–1987

    2023

  15. [23]

    Yuyuan Li, Chaochao Chen, Yizhao Zhang, Weiming Liu, Lingjuan Lyu, Xiaolin Zheng, Dan Meng, and Jun Wang. 2023. Ultrare: Enhancing receraser for recom- mendation unlearning via error decomposition.Advances in Neural Information Processing Systems36 (2023), 12611–12625

    2023

  16. [24]

    Zihao Li, Dongqi Fu, and Jingrui He. 2023. Everything evolves in personalized pagerank. InProceedings of the ACM Web Conference 2023. 3342–3352

    2023

  17. [25]

    Wenyan Liu, Juncheng Wan, Xiaoling Wang, Weinan Zhang, Dell Zhang, and Hang Li. 2022. Forgetting fast in recommender systems.arXiv preprint arXiv:2208.06875(2022)

    2022 arXiv

  18. [26]

    Kevin Meng, David Bau, Alex Andonian, and Yonatan Belinkov. 2022. Locating and editing factual associations in gpt.Advances in neural information processing systems35 (2022), 17359–17372

    2022

  19. [27]

    Gaurav Patel and Qiang Qiu. 2025. Learning to unlearn while retaining: Com- bating gradient conflicts in machine unlearning. InProceedings of the IEEE/CVF International Conference on Computer Vision. 4211–4221

    2025

  20. [28]

    Hadi Reisizadeh, Jinghan Jia, Zhiqi Bu, Bhanukiran Vinzamuri, Anil Ramakr- ishna, Kai-Wei Chang, Volkan Cevher, Sijia Liu, and Mingyi Hong. 2025. BLUR: A Bi-Level Optimization Approach for LLM Unlearning.arXiv preprint arXiv:2506.08164(2025)

    2025

  21. [29]

    Guangyuan Shi, Qimai Li, Wenlong Zhang, Jiaxin Chen, and Xiao-Ming Wu

  22. [30]

    arXiv preprint arXiv:2302.11289(2023)

    Recon: Reducing conflicting gradients from the root for multi-task learning. arXiv preprint arXiv:2302.11289(2023)

    2023 arXiv

  23. [31]

    Aaquib Syed, Can Rager, and Arthur Conmy. 2024. Attribution patching out- performs automated circuit discovery. InProceedings of the 7th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP. 407–416

    2024

  24. [32]

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yas- mine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhos- ale, et al. 2023. Llama 2: Open foundation and fine-tuned chat models.arXiv preprint arXiv:2307.09288(2023)

    2023 arXiv

  25. [33]

    Hangyu Wang, Jianghao Lin, Bo Chen, Yang Yang, Ruiming Tang, Weinan Zhang, and Yong Yu. 2025. Towards efficient and effective unlearning of large language models for recommendation.Frontiers of Computer Science19, 3 (2025), 193327

    2025

  26. [34]

    Kevin Wang, Alexandre Variengien, Arthur Conmy, Buck Shlegeris, and Jacob Steinhardt. 2022. Interpretability in the wild: a circuit for indirect object identifi- cation in gpt-2 small.arXiv preprint arXiv:2211.00593(2022)

    2022 arXiv

  27. [35]

    Peng Wang, Zexi Li, Ningyu Zhang, Ziwen Xu, Yunzhi Yao, Yong Jiang, Pengjun Xie, Fei Huang, and Huajun Chen. 2024. Wise: Rethinking the knowledge memory for lifelong model editing of large language models.Advances in Neural Information Processing Systems37 (2024), 53764–53797

    2024

  28. [36]

    Mingji Yang, Hanzhi Wang, Zhewei Wei, Sibo Wang, and Ji-Rong Wen. 2024. Efficient algorithms for personalized pagerank computation: A survey.IEEE Transactions on Knowledge and Data Engineering36, 9 (2024), 4582–4602

    2024

  29. [37]

    Yuanshun Yao, Xiaojun Xu, and Yang Liu. 2024. Large language model unlearning. Advances in Neural Information Processing Systems37 (2024), 105425–105475

    2024

  30. [38]

    Biao Yi, Jiahao Li, Baolei Zhang, Lihai Nie, Tong Li, Tiansheng Huang, and Zheli Liu. 2025. Gradient surgery for safe llm fine-tuning.arXiv preprint arXiv:2508.07172(2025)

    2025 arXiv

  31. [39]

    Tianhe Yu, Saurabh Kumar, Abhishek Gupta, Sergey Levine, Karol Hausman, and Chelsea Finn. 2020. Gradient surgery for multi-task learning.Advances in neural information processing systems33 (2020), 5824–5836

    2020

  32. [40]

    Dan Zhang, Tao Feng, Lilong Xue, Yuandong Wang, Yuxiao Dong, and Jie Tang

  33. [41]

    Parameter-efficient fine-tuning for foundation models.arXiv preprint arXiv:2501.13787(2025)

    2025 arXiv

  34. [42]

    Fred Zhang and Neel Nanda. 2023. Towards best practices of activation patching in language models: Metrics and methods.arXiv preprint arXiv:2309.16042(2023)

    2023 arXiv

  35. [43]

    Xingyi Zhang, Zixuan Weng, and Sibo Wang. 2024. Towards deeper understand- ing of ppr-based embedding approaches: a topological perspective. InProceedings of the ACM Web Conference 2024. 969–979

    2024

Pith tools

Reviewed July 13, 2026 · model on record in the stance chip above.