Unveiling Biological Models Through Turing Patterns

Catharine W. K. Lo; Hongyu Liu; Yuhan Li

REVIEW 4 major objections 4 minor 23 references

Unveiling Biological Models Through Turing Patterns

T0 review · 4 major / 4 minor · reviewed 2026-08-04 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Fourier amplitudes alone determine all model parameters from a single Turing pattern.

desk verdict Fresh idea — using Turing pattern amplitudes as inverse data — but the proof is a Galerkin handwave, not a theorem. read the letter →

arxiv 2509.07458 v1 pith:EAF6W6I2 submitted 2025-09-09 math.AP physics.bio-phq-bio.BMq-bio.CB

Yuhan Li , Hongyu Liu , Catharine W. K. Lo This is my paper · ORCID

classification math.APphysics.bio-phq-bio.BMq-bio.CB MSC 35R3035B1035B3635K1035K5535K5735Q9292-10

keywords Turingpatternsinverseproblemparameteridentificationchemotaxisreaction-diffusionsystemsFourieramplitudesstationaryperiodicsolutionswavenumberrecovery

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

Stationary Turing patterns carry enough information in their shape alone to name the mechanism that made them. The paper proves that, for two chemotaxis models—one with density-dependent chemotaxis and one with ratio-dependent chemotaxis—the Fourier cosine amplitudes of a single periodic pattern uniquely determine the two diffusion coefficients, the chemotactic strength, the logistic growth parameter, and the spatial wavelength. The proof works by substituting a truncated cosine expansion of the pattern into the stationary reaction-diffusion equations and equating coefficients mode by mode; at truncation order three (or two for the second model) the resulting algebraic system has as many independent equations as unknowns. If the claim holds, biologists could in principle read the mechanistic parameters of a patterning system directly from one spatial image of the pattern, without time-series or boundary measurements. The authors present this as a new direction for inverse problems in biology, taking the pattern itself as the observable.

What carries the argument

The carrying object is the truncated Fourier cosine representation of the pattern on the no-flux interval [0,L], together with the stationary PDE obtained by setting time derivatives to zero. Plugging the series into the PDE and using orthogonality of cosines converts the PDE into a finite algebraic system whose unknowns are d_n, d_c, χ0, r, and k. For Model 1 the decisive system is (3.20), five equations at truncation M=3; for Model 2 it is the M=1 system (4.9) together with (4.4), or the M=2 system (4.22)-(4.26), which requires first clearing rational denominators. The systems are designed to have as many independent equations as parameters, which is the entire mechanism of the uniqueness

What would settle it

Simulate model (1.4) with known parameter values on a fine grid, extract the Fourier amplitudes of the stationary pattern, truncate at M=3, solve system (3.20), and compare the recovered d_n, d_c, χ0, r, k with the true values. If the recovered values do not approach the true ones as the simulation grid and truncation order are refined, the exact-truncation premise collapses. A second check: find amplitude data for which the five equations (3.20) have more than one solution, which would directly refute the uniqueness claim.

Watch

Extended reading notes

Core claim

At the center of the paper is a parameter-counting fact. The authors write a stationary one-dimensional Turing pattern from either chemotaxis model (1.4) or (1.5) as a truncated cosine series whose amplitudes α_i are the measured data. Substituting that series into the stationary reaction-diffusion equations, applying trigonometric product-to-sum identities, and equating the coefficients of each cosine mode yields a system of algebraic equations. For Model 1, the five equations obtained at truncation order M=3 are claimed to be linearly independent in the five unknowns d_n, d_c, χ0, r, k; for Model 2, the equations obtained at M=1 (with the β1 relation) or at M=2 serve the same role. Theorem

Load-bearing premise

The proof assumes that a real pattern is exactly captured by the first few cosine modes used in the calculation, so that dropping all higher modes loses nothing; if it does lose something, the recovered parameters may not be the true ones.

Editorial extensions

If this is right

A single spatial snapshot of a mature pattern is claimed to be sufficient input; the recovery does not need boundary measurements, time-series data, or multiple experiments.
The recovered wavelength and diffusion coefficients come from the same equations, so the pattern's observed spatial scale directly determines transport parameters.
The same Fourier-coefficient substitution applies to any reaction-diffusion-advection model that produces stationary periodic Turing patterns, making the approach a general template rather than a two-model trick.
At truncation order M=3 the cited numerics report a pattern representation error below one percent, so the algebraic recovery is expected to produce quantitatively reliable parameters in practice.
Pattern images from nature, such as coat markings or microbial colony contrasts, become legitimate data for mechanistic model identification.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A stability analysis is not given; the natural next step is to quantify how recovery error scales with truncation order and measurement noise, since the exact-truncation argument may fail for noisy or under-resolved patterns.
The assumed functional forms (for example χ=χ0 n versus χ=χ0 n/c, and f=rn(1−n), g=n−c) are fixed in advance, so amplitude data alone cannot choose between model classes; one could use residual higher harmonics as a model-selection test.
Feeding the same amplitude data through both algebraic pipelines and comparing the recovered parameter sets offers a built-in consistency check that the paper does not explore.
Because the proof assumes the pattern is exactly stationary, a robustness test on slowly drifting or transient patterns would clarify how wide the applicability really is.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Quick take: the idea here is genuinely interesting. Using the spatial amplitudes of a Turing pattern as the data for a parameter-identification problem is a fresh framing, and I know of no direct precedent beyond Kazarnikov et al.'s statistical work. That part is worth taking seriously.

The paper doesn't deliver on its central claim, though. Theorems 2.1 and 2.2 assert unique recovery of all five parameters from a single pattern's Fourier amplitudes. The argument substitutes a truncated M=3 cosine series into the stationary PDE and then equates selected Fourier coefficients. That is a Galerkin projection, not an exact solution method. For the nonlinear logistic term, a finite Fourier series generically produces residual harmonics up to order 2M — here the constant term and cos(6kx) — and those residuals are silently dropped. Nothing in the paper controls the gap between parameters recovered from the projected equations and the true parameters. The claim that the five equations are 'clearly linearly independent' doesn't establish existence, uniqueness, or stability; it's a nonlinear algebraic system, and linear independence of the left-hand sides isn't the right notion anyway.

There are also concrete errors. The sign in the fifth equation of (3.20) is wrong relative to the expansions in (3.17) and (3.19): the term should be −rα2α3, not +. Section 4 gets even more algebra-heavy, and I would not trust the expansions without computer verification. The abstract promises recovery of 'full nonlinear forms' of coefficient functions, but the proof only handles two constant scalars in two toy models. There are no numerical experiments. The forward accuracy result from [5] doesn't tell us anything about conditioning of the inverse map.

So the paper is best read as a research proposal with a worked example, not as a proven theorem. The core idea is legitimate and might be salvageable with (a) a proper convergent-expansion framework with error estimates, or (b) a computational study that actually demonstrates stable recovery on simulated data. As it stands, I would not trust the identifiability result.

My recommendation: this deserves a serious referee — the idea is novel enough that a good referee could help the authors see what a real proof would require. But it should be sent back for major revision, and if the central issue isn't fixed, rejection is right.

Referee Report

4 major / 4 minor

Summary. The paper studies an inverse problem for two chemotaxis reaction-diffusion models, (1.4) and (1.5). The authors assume that a stationary Turing pattern is represented by a truncated Fourier cosine series, Eq. (2.1), and that the Fourier amplitudes {α_i} are the available data. The central claim is Theorem 2.1 (and Theorem 2.2) that from these amplitudes one can uniquely recover the diffusion coefficients, chemotactic coefficient, kinetic parameter, and wavenumber. The proof substitutes the truncated series into the stationary PDEs, expands the nonlinear terms, and equates selected Fourier coefficients to obtain algebraic systems such as (3.20) and (4.22)–(4.26), which are then declared to determine the five unknowns.

Significance. If correct, the result would be a significant methodological advance: a single spatial snapshot of a Turing pattern would determine all parameters of the underlying model, a genuinely new inverse-problem paradigm. The paper also correctly identifies that pattern amplitudes are natural biological observables. However, the central proof is a formal Galerkin argument. The truncation is assumed exact without controlling the residual harmonics; the resulting algebraic systems are not proved to have a unique solution; and at least one displayed equation contains a sign error. No numerical experiments or stability estimates are provided. The main theorems are therefore not established, and in their current form the claims substantially exceed what is proved.

major comments (4)

[§3, Eqs. (2.1)–(3.20)] The proof treats the M=3 truncation in (2.1) as an exact representation of the stationary pattern. Substituting a finite cosine series into the nonlinear PDE (2.5) produces residual harmonics up to order 6 (e.g., cos(6kx) and the constant term), which are silently dropped. The true parameters therefore do not necessarily satisfy the truncated system (3.20). The numerical statement from [5] that the forward truncation has <1% error does not quantify the gap between solutions of the Galerkin system and the true inverse problem; no error bounds or stability estimates are given. This undermines Theorem 2.1 directly.
[§3, Eq. (3.20)] The last equation in (3.20) has a sign error. From the expansions in (3.17) and (3.19), the coefficient of cos(5kx) in the first equation of (2.5) is χ0 k^2(15/2 α2β3 + 5α3β2) − r α2 α3, not + r α2 α3. The displayed equation is therefore inconsistent with the preceding derivation, and the claimed system is not the one actually obtained.
[§3, Eq. (3.20); §4, Eqs. (4.22)–(4.26)] The assertion that the five equations are 'clearly linearly independent' is not a meaningful statement for a nonlinear algebraic system, and in any case linear independence does not imply existence or uniqueness of a solution. The paper neither proves that the true parameters solve the selected finite equations (owing to truncation residuals) nor that a solution of the finite system recovers the true parameters. The argument reduces to solving an ad hoc Galerkin projection, with no quantified relation to the original infinite-dimensional inverse problem.
[§4, Eq. (4.9)] At M=1 for Model 2, the paper states that four equations together with (4.4) determine five unknowns. However, (4.4) is merely the expression β1 = α1/(1 + d_c k^2), i.e., a definition used to eliminate β1. After substitution there are four equations for the five unknowns d_n, d_c, k, χ0, r, so the system is underdetermined. The later passage to M=2 does not repair this logical gap, since the same truncation-exactness issue persists.

minor comments (4)

[Abstract; §1.4; Theorem 1.1] The paper repeatedly claims recovery of the 'full nonlinear forms' of χ, f, and g, but Theorems 2.1 and 2.2 only treat two specific models with constant coefficients. The broad claim is not supported by the results.
[§2.2, Eqs. (2.3)–(2.4)] The symbol M is used both for the truncation order in (2.1) and for the measurement set in (2.3), which is confusing. Rename one of them.
[§3–§4] The derivation contains several typographical slips (e.g., in the long expansions in §4) that make verification difficult. A pass with computer algebra verification or a supplementary notebook would improve reliability.
[General] No numerical examples are given to illustrate the proposed recovery, even for synthetic data generated by the forward models. Such experiments would be necessary to support the practical claims made in the introduction.

Circularity Check

1 steps flagged · score 4.0 of 10

Wavenumber recovery is fixed by the definition of the measurement basis; central coefficient recovery is non-circular but incomplete.

self definitional [Section 2.1 (Eqs. (2.1)-(2.3)) and Section 2.2 (Theorem 2.1)]
"The associated wavelength is given by k=π/L. ... Given a Turing profile, the amplitudes α_i of n(x) can be computed as follows: ... α_i = 2/L ∫_0^L n(x) cos(ikx) dx ... M={α_i}_{i=0}^M. ... Theorem 2.1. ... with associated amplitude measurements M, we can uniquely compute ... the wavelength k characterizing the Turing pattern."

The measurement M is defined by Fourier projection onto the basis cos(ikx), whose fundamental frequency is k=π/L. Therefore k is already fixed by the construction of the measurements: one cannot compute the α_i without knowing k. The theorem's recovery of k from M is thus a restatement of the basis used to define the data, not an independent inversion result. This is circularity in the wavenumber claim only; the recovery of dn, dc, χ0, r is not defined in terms of M by construction and retains independent content.

full rationale

Apart from the wavenumber issue, the paper's derivation is not circular in the sense of the analyzer: the parameters dn, dc, χ0, r are obtained by substituting the cosine representation into the stationary PDEs and solving the finite algebraic system (3.20), rather than by relabeling fitted inputs as predictions. The self-citations in the introduction are not load-bearing, and the numerical truncation claim is imported from an external paper [5], not from the authors' own prior work. However, the measurement M in (2.2)-(2.3) is defined using k=π/L, making the subsequent recovery of k in Theorems 2.1-2.2 tautological. The more serious mathematical weaknesses—treating the M=3 truncated series as exact, asserting that linearly independent algebraic equations guarantee a unique simultaneous solve, ignoring residual harmonics such as the constant and cos(6kx) terms, and a sign error in the fifth equation of (3.20)—are correctness/rigor issues rather than circularity. Section 4 also explicitly omits a cumbersome expansion ('we skip the process and present the result directly'), which is an omitted computation relevant to proof completeness but not to circularity. Hence the moderate score reflects the partial, definitional circularity in the wavelength recovery while the core coefficient identification remains non-circular.

Assumptions & free parameters 1 free parameters · 5 assumptions · 0 invented entities

The paper's central claim rests on the untested assumption that a finite Fourier projection exactly captures the true solution, plus the usual well-posedness assumptions of the forward models. No free parameters are fitted in the sense of the ledger; the unknowns are the targets, but the truncation order is a hand-picked modeling choice.

free parameters (1)

Truncation order M = 3
Chosen by hand, justified only by a forward-simulation accuracy claim from reference [5]; the entire proof depends on the residual vanishing for modes up to 3.

assumptions (5)

ad hoc to paper The stationary Turing pattern is exactly representable by the truncated Fourier series (2.1) with M=3, so that substituting into the PDE and equating coefficients is exact.
Invoked in Section 3 for equations (3.20); no error bound is derived.
domain assumption The residual of the PDE is orthogonal to the first M+1 cosine modes (Galerkin condition), i.e., the exact solution satisfies the finite projected system.
Standard numerical projection; requires the exact solution to lie in the finite-dimensional space for exact identification, which is not true in general.
domain assumption Model coefficients are piecewise constant on the pattern domain.
Stated in Section 2.2 to reduce functional unknowns to scalars; not derived from biology.
domain assumption The models (1.4) and (1.5) admit stationary periodic Turing patterns for the parameters being recovered; existence is taken from forward literature [5].
The paper does not prove existence of such stationary solutions for the relevant parameters.
domain assumption n and c remain positive so that the ratio n/c in Model 2 is well-defined.
Mentioned in Section 4 where denominators are said to be positive; no proof of positivity.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Unveiling Biological Models Through Turing Patterns." pith.science (2026). https://pith.science/paper/EAF6W6I2

@misc{pith2026250907458,
  author       = {Pith},
  title        = {Pith review of: Unveiling Biological Models Through Turing Patterns},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EAF6W6I2}},
  note         = {Machine review of arXiv:2509.07458}
}

read the original abstract

Turing patterns play a fundamental role in morphogenesis and population dynamics, encoding key information about the underlying biological mechanisms. Yet, traditional inverse problems have largely relied on non-biological data such as boundary measurements, neglecting the rich information embedded in the patterns themselves. Here we introduce a new research direction that directly leverages physical observables from nature--the amplitude of Turing patterns--to achieve complete parameter identification. We present a framework that uses the spatial amplitude profile of a single pattern to simultaneously recover all system parameters, including wavelength, diffusion constants, and the full nonlinear forms of chemotactic and kinetic coefficient functions. Demonstrated on models of chemotactic bacteria, this amplitude-based approach establishes a biologically grounded, mathematically rigorous paradigm for reverse-engineering pattern formation mechanisms across diverse biological systems.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 23 canonical work pages

[5]

Modelling formation of stationary periodic patterns in growing population of motile bacteria

Valentina Bucur and Bakhtier Vasiev. Modelling formation of stationary periodic patterns in growing population of motile bacteria.arXiv preprint arXiv:2406.07182, 2024

work page Pith review arXiv 2024
[1]

Philip Ball. Forging patterns and making waves from biology to geology: a commentary on Turing (1952) ‘The chemical basis of morphogenesis’.Philosophical Transactions of the Royal Society B: Biological Sciences, 370(1666):20140218, 2015

work page 1952
[2]

A model for generating aspects of zebra and other mammalian coat patterns.Journal of Theoretical Biology, 93(2):363–385, 1981

Jonathan BL Bard. A model for generating aspects of zebra and other mammalian coat patterns.Journal of Theoretical Biology, 93(2):363–385, 1981

work page 1981
[3]

Predataxis behavior in Myxococcus xanthus.Proceedings of the National Academy of Sciences, 105(44):17127–17132, 2008

James E Berleman, Jodie Scott, Tatiana Chumley, and John R Kirby. Predataxis behavior in Myxococcus xanthus.Proceedings of the National Academy of Sciences, 105(44):17127–17132, 2008

work page 2008
[4]

Myxococcus xanthus chemotaxis homologs DifD and DifG negatively regulate fibril polysaccharide production.Journal of bacteriology, 186(4):1001–1008, 2004

Wesley P Black and Zhaomin Yang. Myxococcus xanthus chemotaxis homologs DifD and DifG negatively regulate fibril polysaccharide production.Journal of bacteriology, 186(4):1001–1008, 2004

work page 2004
[6]

A non-linear analysis of Turing pattern formation.PloS one, 14(8):e0220994, 2019

Yanyan Chen and Javier Buceta. A non-linear analysis of Turing pattern formation.PloS one, 14(8):e0220994, 2019

work page 2019
[7]

Determining Sources in the Bioluminescence Tomography Problem

Ming-Hui Ding, Rongfang Gong, Hongyu Liu, and Catharine WK Lo. Determining sources in the bioluminescence tomography problem.arXiv preprint arXiv:2311.05191, 2023. 21

work page Pith review arXiv 2023
[8]

Inverse problems for coupled nonlocal nonlinear systems arising in mathematical biology

Ming-Hui Ding, Hongyu Liu, and Catharine WK Lo. Inverse problems for coupled nonlocal nonlinear systems arising in mathematical biology.arXiv preprint arXiv:2407.15713, 2024

work page Pith review arXiv 2024

Show all 23 references

[9]

Determining internal topological struc- tures and running cost of mean field games with partial boundary measurement.arXiv: 2408.08911, 2024

Ming-Hui Ding, Hongyu Liu, and Guang-Hui Zheng. Determining internal topological struc- tures and running cost of mean field games with partial boundary measurement.arXiv: 2408.08911, 2024

2024 arXiv
[10]

Linear stability analysis of high-and low-dimensional models for describing mixing-limited pattern formation in homogeneous autocatalytic reactors

Ankur Gupta and Saikat Chakraborty. Linear stability analysis of high-and low-dimensional models for describing mixing-limited pattern formation in homogeneous autocatalytic reactors. Chemical Engineering Journal, 145(3):399–411, 2009

2009
[11]

A user’s guide to PDE models for chemotaxis.Journal of Mathematical Biologyy, 58(1):183–217, 2009

Thomas Hillen and Kevin J Painter. A user’s guide to PDE models for chemotaxis.Journal of Mathematical Biologyy, 58(1):183–217, 2009

2009
[12]

Human identification and fingerprints: a review.Journal of Biometrics and Biostatistics, 2(123):2, 2011

Nitin Kaushal and Purnima Kaushal. Human identification and fingerprints: a review.Journal of Biometrics and Biostatistics, 2(123):2, 2011

2011
[13]

Statistical approach for parameter identification by Turing patterns.Journal of Theoretical Biology, 501:110319, 2020

Alexey Kazarnikov and Heikki Haario. Statistical approach for parameter identification by Turing patterns.Journal of Theoretical Biology, 501:110319, 2020

2020
[14]

Initiation of slime mold aggregation viewed as an instability

Evelyn F Keller and Lee A Segel. Initiation of slime mold aggregation viewed as an instability. Journal of Theoretical Biology, 26(3):399–415, 1970

1970
[15]

On inverse problems in predator-prey models

Yuhan Li, Hongyu Liu, and Catharine WK Lo. On inverse problems in predator-prey models. Journal of Differential Equations, 397:349–376, 2024

2024
[16]

Determining a parabolic-elliptic-elliptic system by boundary observation of its non-negative solutions under chemotaxis background.arXiv preprint arXiv:2509.04850, 2025

Yuhan Li, Hongyu Liu, and Catharine WK Lo. Determining a parabolic-elliptic-elliptic system by boundary observation of its non-negative solutions under chemotaxis background.arXiv preprint arXiv:2509.04850, 2025

2025 arXiv
[17]

On inverse problems in multi-population aggregation models.Journal of Differential Equations, 414:94–124, 2025

Yuhan Li, Hongyu Liu, and Catharine WK Lo. On inverse problems in multi-population aggregation models.Journal of Differential Equations, 414:94–124, 2025

2025
[18]

On the simultaneous recovery of environmental factors in the 3D chemotaxis-Navier-Stokes models.Communications on Analysis and Computation, 2(1):30–47, 2024

Yuhan Li and Catharine WK Lo. On the simultaneous recovery of environmental factors in the 3D chemotaxis-Navier-Stokes models.Communications on Analysis and Computation, 2(1):30–47, 2024

2024
[19]

Determining a parabolic system by boundary observation of its non-negative solutions with biological applications.Inverse Problems, 40(2):025009, 2024

Hongyu Liu and Catharine WK Lo. Determining a parabolic system by boundary observation of its non-negative solutions with biological applications.Inverse Problems, 40(2):025009, 2024

2024
[20]

Springer, 2003

James D Murray.Mathematical biology: II: spatial models and biomedical applications, vol- ume 18. Springer, 2003

2003
[21]

An introduction, volume 17

James D Murray.Mathematical biology: I. An introduction, volume 17. Springer Science & Business Media, 2007

2007
[22]

Cambridge University Press, 1989

Taylor A Steeves and Ian M Sussex.Patterns in plant development. Cambridge University Press, 1989

1989
[23]

The chemical basis of morphogenesis.Bulletin of Mathematical Biology, 52:153–197, 1990

Alan M Turing. The chemical basis of morphogenesis.Bulletin of Mathematical Biology, 52:153–197, 1990. 22

1990

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.

[1] [5]

Modelling formation of stationary periodic patterns in growing population of motile bacteria

Valentina Bucur and Bakhtier Vasiev. Modelling formation of stationary periodic patterns in growing population of motile bacteria.arXiv preprint arXiv:2406.07182, 2024

work page Pith review arXiv 2024

[2] [1]

Philip Ball. Forging patterns and making waves from biology to geology: a commentary on Turing (1952) ‘The chemical basis of morphogenesis’.Philosophical Transactions of the Royal Society B: Biological Sciences, 370(1666):20140218, 2015

work page 1952

[3] [2]

A model for generating aspects of zebra and other mammalian coat patterns.Journal of Theoretical Biology, 93(2):363–385, 1981

Jonathan BL Bard. A model for generating aspects of zebra and other mammalian coat patterns.Journal of Theoretical Biology, 93(2):363–385, 1981

work page 1981

[4] [3]

Predataxis behavior in Myxococcus xanthus.Proceedings of the National Academy of Sciences, 105(44):17127–17132, 2008

James E Berleman, Jodie Scott, Tatiana Chumley, and John R Kirby. Predataxis behavior in Myxococcus xanthus.Proceedings of the National Academy of Sciences, 105(44):17127–17132, 2008

work page 2008

[5] [4]

Myxococcus xanthus chemotaxis homologs DifD and DifG negatively regulate fibril polysaccharide production.Journal of bacteriology, 186(4):1001–1008, 2004

Wesley P Black and Zhaomin Yang. Myxococcus xanthus chemotaxis homologs DifD and DifG negatively regulate fibril polysaccharide production.Journal of bacteriology, 186(4):1001–1008, 2004

work page 2004

[6] [6]

A non-linear analysis of Turing pattern formation.PloS one, 14(8):e0220994, 2019

Yanyan Chen and Javier Buceta. A non-linear analysis of Turing pattern formation.PloS one, 14(8):e0220994, 2019

work page 2019

[7] [7]

Determining Sources in the Bioluminescence Tomography Problem

Ming-Hui Ding, Rongfang Gong, Hongyu Liu, and Catharine WK Lo. Determining sources in the bioluminescence tomography problem.arXiv preprint arXiv:2311.05191, 2023. 21

work page Pith review arXiv 2023

[8] [8]

Inverse problems for coupled nonlocal nonlinear systems arising in mathematical biology

Ming-Hui Ding, Hongyu Liu, and Catharine WK Lo. Inverse problems for coupled nonlocal nonlinear systems arising in mathematical biology.arXiv preprint arXiv:2407.15713, 2024

work page Pith review arXiv 2024