Application of continuous corrections in widefield imaging obtained with the IDGCAL calibration method

Sebastiaan van der Tol

REVIEW 3 major objections 6 minor 17 references

Application of continuous corrections in widefield imaging obtained with the IDGCAL calibration method

T0 review · 3 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read The paper argues that reordering the derivative loop and switching to Karhunen-Loève basis functions will make the IDG-CAL calibration method feasible for deeper wide-field radio images.

desk verdict A sound reordering trick plus a promising but unverified KL-basis proposal, marred by a typo in Eq. (11) and an admitted stand-in stochastic model. read the letter →

arxiv 2608.10762 v1 pith:7QA6CZWO submitted 2026-08-11 astro-ph.IM

Sebastiaan van der Tol This is my paper · ORCID

classification astro-ph.IM

keywords radiointerferometrywide-fieldimagingdirection-dependentcalibrationimage-domaingriddingA-projectionKarhunen-LoèvebasisfunctionsIDG-CAL

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

This paper argues that two changes will make IDG-CAL, a calibration method for direction-dependent effects in wide-field radio interferometry, practical for deep images. First, the derivative computation can be reordered so the loop over free parameters is moved outside the inner loop, meaning runtime no longer needs to scale linearly with the number of parameters. Second, the polynomial basis functions currently used to describe the A-terms should be replaced by Karhunen-Loève basis functions derived from an assumed stochastic model of gain fluctuations, optionally weighted by the model image so the basis reserves detail at bright sources. The paper provides the derivations for both improvements and states that implementation is still pending. A sympathetic reader would care because IDG-CAL could then produce smooth, physically plausible corrections over the full field at a cost comparable to current facet-based methods.

What carries the argument

The two load-bearing mechanisms are the reordered derivative expression in Eq. (8) and the Karhunen-Loève construction of Eqs. (11)–(15). Equation (8) rewrites the per-parameter derivative of the cost function as $-2 \mathrm{Re} \sum_{q=1}^{Q} \operatorname{vec}(Y_{sq})^{H} \operatorname{vec}(\partial A_{sq}^{H}/\partial x_{sr})$, where $Y_{sq}$ accumulates sub-grid contributions per pixel before any parameter loop, removing the linear-in-parameters scaling for the complex diagonal terms. The basis-function mechanism builds a correlation matrix from the assumed structure function $\mathbb{E}[(x(p_1)-x(p_2))^2]=\|p_1-p_2\|^\beta$, projects out the constant mode, takes an eigenvalue decomposition to get the KL basis functions, and optionally weights by the model-image flux, followed by a QR-decomposition to restore orthonormality.

What would settle it

Measure the actual structure function of the A-term phase and amplitude from calibration solutions on the full dataset; if it is not a power law over the field separations used, the Karhunen-Loève basis built from Eq. (11) is not optimal and the claimed quality gain should not appear in the residual images.

Watch

Extended reading notes

Core claim

The central claim is that IDG-CAL's barriers to deeper imaging, runtime linear in the number of free parameters and polynomials that are steep at the field edges, are removable by a computational reordering and a better basis. Equation (8) rewrites the derivative of the cost function as a sum over accumulated sub-grid pixels, with the parameter derivative factored outside the sub-grid sum, so the costly loop over parameters is no longer in the inner loop. The second proposal constructs the A-term expansion from the eigenfunctions, the Karhunen-Loève functions, of a correlation matrix built from a Kolmogorov structure function, with a constant term added back; weighting that correlation matrix by the model image flux and then undoing the weight via a QR-decomposition yields orthonormal basis functions that show more detail at bright sources. The paper presents this as a derivation and a proposal, not yet as an implementation.

Load-bearing premise

The improved basis functions are optimal only if the gain fluctuations really follow the assumed turbulence power law, and the paper itself notes that this model is not an accurate description of telescope-based effects.

Editorial extensions

If this is right

Runtime for A-terms with many parameters no longer scales linearly with the number of free parameters, making higher-order screens feasible on full-bandwidth datasets.
A-terms described by the KL basis stay smooth across the field while concentrating resolving power at the brighter sources, avoiding the edge steepness of high-order polynomials.
Because the flux weighting can be averaged per station, station-specific basis functions can be constructed without changing the derivation.
If the implementation that is still pending works as derived, IDG-CAL could match or beat facet-based calibration at full bandwidth with a smaller number of smooth coefficients.

Reading between the lines

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

Inference: the loop-reordering trick is general, so any calibration or imaging step where a per-parameter derivative factors out of an accumulated sum could receive the same speed-up.
Inference: since the paper concedes that the Kolmogorov model is a stand-in for instrument-based effects, a natural extension is to estimate the structure function from real calibration solutions and rebuild the basis functions from that measured correlation matrix.
Inference: the flux-weighted basis is a tunable middle ground between fully independent per-source gains and a single smooth screen; varying the weighting could trade source isolation against global smoothness in ways the paper leaves unexplored.
Inference: if the runtime saving is as large as the loop nesting suggests, the practical limit on A-term order may shift from compute time to conditioning of the least-squares problem, which the paper does not discuss.

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.

Desk Editor's Note

read the letter

Colleague,

This is a short, honest methods proposal, not a demonstrated result. The first improvement—reordering the derivative computation so the parameter loop is outside the inner loop—is the stronger of the two. The vectorization identity in Eqs. (3)–(8) checks out, and the idea of accumulating sub-grids before looping over basis coefficients is a clean way to decouple cost from the number of free parameters, at least for the long-timescale diagonal terms. That alone is worth seeing in print.

The second improvement, KL basis functions, is more of a mixed bag. Constructing A-term basis functions from a stochastic model of gain fluctuations is genuinely new in this subfield, and the flux-weighted variant (Eqs. 14–15) is a sensible way to concentrate degrees of freedom at bright sources. But the derivation has a real typo: Eq. (11) uses ((l1−l2)^2 (m1−m2)^2)^{β/2}, which is a product, not the Kolmogorov structure function it claims to encode. The correct form should have a plus sign. As printed, the matrix is not the correlation function described in Eq. (9), and the derivation is not reproducible. Also, the 'optimality' is conditional on a power-law structure function with an unspecified β, and the paper itself concedes that model is not accurate for instrumental effects. No sensitivity test is offered. So the second improvement is an idea with an admitted stand-in model and a typo in its key equation.

The paper is honest about its status: it reports that the previous implementation didn't scale, and it ends with 'implementation is still need to be done.' No timing or imaging results are presented, so the title's 'application' overstates what is actually in the paper. But as a proposal, the reasoning is clear and the limitations are acknowledged; the citation pattern is appropriate.

Who gets value? Someone working on DDE calibration for wide-field surveys will find the reordering derivation and the KL-basis construction worth studying, even if they won't adopt it as-is. It deserves a serious referee, but the referee should demand the typo fix, a discussion of β or a robustness test, and language that clearly frames the second part as a proposal rather than a proven improvement.

Recommendation: accept for peer review, with expected heavy revision.

Referee Report

3 major / 6 minor

Summary. The paper reports on the ongoing development of IDG-CAL, an image-domain calibration method for direction-dependent effects in radio interferometry. It describes the current status of the algorithm on a full-scale dataset and proposes two improvements: (1) reordering the derivative computation so that the loop over free parameters is moved outside the inner loop, and (2) replacing polynomial basis functions with Karhunen-Loève basis functions derived from a stochastic model of gain fluctuations. The reordering derivation is presented in Eqs. (3)-(8), and the KL basis construction is presented in Eqs. (9)-(15). The paper states that these improvements are needed to make deep wide-field imaging feasible, but it also states explicitly in the conclusion that the implementation still needs to be done.

Significance. If the proposed improvements are realized, they could address genuine bottlenecks in IDG-CAL: the linear scaling of runtime with the number of free parameters and the poor behavior of high-order polynomial basis functions. The algebraic reordering in Section 4.1 is a useful and apparently correct step that could reduce computational cost for high-dimensional A-term descriptions. The KL basis construction is a principled alternative to polynomials and the flux-weighting idea in Eqs. (14)-(15) is sensible. However, the manuscript as it stands is a proposal: the central claims that these improvements will make deep imaging feasible and that the derived basis functions are optimal are not backed by any implementation, benchmark, or data-driven validation. The paper's value is therefore conditional on future work.

major comments (3)

[§4.2, Eq. (11)] The correlation matrix in Eq. (11) is printed with off-diagonal entries of the form ((l1-l2)^2 (m1-m2)^2)^{β/2}, i.e., the product of the squared coordinate differences. The structure function in Eq. (9) is ||p1-p2||^β, which for coordinates (l,m) corresponds to a Euclidean distance, that is ((l1-l2)^2 + (m1-m2)^2)^{β/2}. As written, the matrix is not the claimed Kolmogorov structure function and the subsequent eigenvalue decomposition is not reproducible. This is a load-bearing typo because the entire KL basis construction rests on this matrix.
[§4.2, Eqs. (9)-(15)] The claimed optimality of the KL basis functions is conditional on the stochastic model in Eq. (9), which contains an unspecified exponent β and is explicitly conceded in Section 4.2 to be 'not an accurate description of the physics' for instrument-based effects. The paper gives no procedure for choosing β, no empirical fit to actual gain variations, and no sensitivity analysis. Without such support, the assertion that these basis functions are 'optimal' for the real gain screen is not established; the text itself only says the functions 'might still be good enough,' which is an admission of uncertainty rather than a demonstrated result.
[§4.1 and §5] The paper's central feasibility claim—that with more parameters and the proposed improvements deep imaging becomes practical—is unverified. Section 4.1 states that the reordering 'greatly reduces' the contribution of the parameter loop, but no complexity analysis, operation count, or timing measurement is given. The conclusion states that 'The implementation is still need to be done.' Thus neither the runtime improvement nor the quality improvement is demonstrated, and the reader cannot judge whether the proposed changes will actually make the full-dataset application feasible.

minor comments (6)

[§1 and throughout] The text contains several typographical errors and inconsistencies, e.g., 'the the gridding kernel' in Section 1, 'faced based' instead of 'facet based' in several places, and 'base functions' instead of 'basis functions' in Figure captions and elsewhere.
[§2, Eq. (1)] The expression for J(p) in Eq. (1) is typeset in a garbled manner with an unexplained superscript 'j' and a diagonal matrix whose entries are not fully legible. Please clarify the notation and ensure all factors are correctly written.
[§4.1, Eq. (3)] The sum over q in Eq. (3) is confusing: V_pq is described as a sub-grid pixel, but vec(V_pq) is used inside the sum. Please define V_pq consistently (scalar or matrix) and clarify the meaning of the q-sum in this equation.
[§4.2, Eqs. (12)-(15)] The symbol Q is overloaded: it denotes the number of subgrid pixels in Eq. (3) and is also used as the orthogonal matrix in the QR-decomposition of Eq. (15). Please use distinct symbols to avoid ambiguity.
[§4.2, Eq. (9)] The notation ||p1 - p2|| in Eq. (9) is not explicitly defined for the 2D (l,m) coordinates; please state that this is the Euclidean norm in the image plane, which is important for interpreting the matrix in Eq. (11).
[§5] The sentence 'The implementation is still need to be done' contains a grammatical error; please rephrase, e.g., 'The implementation still needs to be done.'

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reordered derivative is a mathematical rearrangement of prior published equations, and the KL basis is derived from an explicit stochastic model; the admitted model mismatch and the Eq. (11) typo are limitations, not circular steps.

full rationale

The first improvement (Section 4.1) starts from Eq. (7) and Eq. (14) of Ref. 1 and applies the identity vec(ABC)=(C^T⊗A)vec(B) to move the sub-grid accumulation outside the parameter loop, culminating in Eq. (8). This is a mathematical rearrangement of previously published equations, not a fit or a self-citation used to forbid alternatives; using Ref. 1 as the starting point is normal background citation. The second improvement (Section 4.2) constructs Karhunen-Loève basis functions from an explicit stochastic model: Eq. (9) posits a Kolmogorov structure function, Eq. (11) builds the correlation matrix, and Eqs. (12)-(15) perform projection, eigendecomposition, and flux-weighted orthogonalization. The basis functions are, by construction, the eigenvectors of the assumed correlation matrix; this is a derivation from stated assumptions, not a circular prediction, because the model parameters are not fitted from the data whose calibration the basis functions are meant to improve. The paper honestly flags a limitation: for instrumental effects 'this model is not an accurate description of the physics' (Section 4.2), and Eq. (11) as printed appears to contain a typo (the product of squared differences instead of their sum), which is a reproducibility concern rather than a circularity. The implementation and tests are postponed, so the paper makes no empirical claim that would be forced by its own inputs. Overall circularity score: 0.

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

The central proposal rests on a stochastic model (Kolmogorov turbulence) with an unspecified exponent, and on a standard linear algebra identity. No new physical entities are introduced.

free parameters (1)

β (Kolmogorov exponent) = not specified
Appears in the structure function in Eq. (9); the paper never sets it to the usual Kolmogorov value (5/3) nor fits it to data. The resulting basis functions depend on its value.

assumptions (5)

domain assumption The cost function is the mean squared residual visibilities
Used in Section 2 and in the derivative derivation (Eq. 3). This is standard for calibration.
domain assumption Kolmogorov turbulence describes the gain variations
Assumed in Section 4.2, Eq. (9). The paper says this is applicable to atmospheric effects but not instrument-based effects.
ad hoc to paper The autocorrelation of the phase process can be ignored and the constant mode projected out
Stated in Section 4.2: 'Here we ignore the auto correlations, and project the constant term out of the correlation matrix.' This handles the infinite variance of a structure function without outer scale.
standard math vec(ABC) = (C^T ⊗ A) vec(B)
Used in Eq. (4). This is a standard linear algebra identity.
standard math The KL basis functions are optimal in mean-squared error for a Gaussian process with that covariance
Implied in Section 4.2, not proven in the paper. This is a known result of Karhunen-Loève theory.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Application of continuous corrections in widefield imaging obtained with the IDGCAL calibration method." pith.science (2026). https://pith.science/paper/7QA6CZWO

@misc{pith2026260810762,
  author       = {Pith},
  title        = {Pith review of: Application of continuous corrections in widefield imaging obtained with the IDGCAL calibration method},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7QA6CZWO}},
  note         = {Machine review of arXiv:2608.10762}
}

read the original abstract

Image Domain Gridding (IDG) is an efficient method to evaluate A-projection, a method for correcting for direction dependent effects (DDEs) in radio astronomical imaging. The corrections i.e. the A-terms need to be estimated from the observed data through calibration. IDG-CAL is a calibration algorithm that employs the IDG algorithm in an iterative fashion to estimate A-terms directly. Initial results for low order A-terms are promising, however for deeper images higher order A-terms are needed. Here two improvements to IDG-CAL are proposed to 1) reduce the computational cost of A-terms described by more parameters and 2) optimize the set of basis functions describing the A-terms based on a stochastic model of the gain variations and the model image.

Figures

Figures reproduced from arXiv: 2608.10762 by the authors.

**Figure 2.** Overview of IDG predict. The procedure consists of the following steps 1) the model image is inverse Fourier [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** 2D Karhunen-Lo`eve basis functions over ( [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: 2D Karhunen-Lo`eve basis functions over ( [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

17 extracted references · 6 canonical work pages

[1]

, keywords =

Direction-dependent calibration with image-domain gridding. , keywords =. doi:10.1051/0004-6361/202558272 , archivePrefix =. 2602.06002 , primaryClass =

work page arXiv
[2]

Ionospheric calibration of low frequency radio interferometric observations using the peeling scheme. I. Method description and first results. , keywords =. doi:10.1051/0004-6361/200811094 , archivePrefix =. 0904.3975 , primaryClass =

arXiv
[3]

Radio Science , keywords =

Scintillations due to a concentrated layer with a power-law turbulence spectrum. Radio Science , keywords =. doi:10.1029/RS010i001p00107 , adsurl =

work page doi:10.1029/rs010i001p00107
[4]

Correcting direction-dependent gains in the deconvolution of radio interferometric images

Correcting direction-dependent gains in the deconvolution of radio interferometric images. , keywords =. doi:10.1051/0004-6361:20079284 , archivePrefix =. 0805.0834 , primaryClass =

work page Pith review arXiv
[5]

, keywords =

LOFAR Facet Calibration. , keywords =. doi:10.3847/0067-0049/223/1/2 , archivePrefix =. 1601.05422 , primaryClass =

arXiv
[6]

Entropy Encoding, Hilbert Space and Karhunen-Loeve Transforms

Entropy encoding, Hilbert space, and Karhunen-Lo \`e ve transforms. Journal of Mathematical Physics , keywords =. doi:10.1063/1.2793569 , archivePrefix =. math-ph/0701056 , primaryClass =

work page Pith review arXiv
[7]

, TITLE=

Guio, Patrick and Pécseli, Hans L. , TITLE=. Frontiers in Astronomy and Space Sciences , VOLUME=. 2021 , URL=. doi:10.3389/fspas.2020.573746 , ISSN=

work page arXiv 2021
[8]

Ionospheric Calibration for the LOFAR Radio Telescope , year=

van der Tol, Sebastiaan and van der Veen, Alle-Jan , booktitle=. Ionospheric Calibration for the LOFAR Radio Telescope , year=

work page

Show all 17 references

[9]

G. J. Alred and C. T. Brusaw and W. E. Oliu. Handbook of Technical Writing. 2015 (eleventh edition)

2015
[10]

The LaTeX Companion

Michel Goossens and Frank Mittelbach and Sebastian Rahtz. The LaTeX Companion. 1997

1997
[11]

The LaTeX Companion

Frank Mittelbach and Michel Goossens and Johannes Braams and David Carlisle. The LaTeX Companion. 2004

2004
[12]

S. F. Gull. Developments in maximum-entropy data analysis. Maximum Entropy and Bayesian Methods. 1989

1989
[13]

K. M. Hanson. Introduction to B ayesian image analysis. Medical Imaging:\ Image Processing. 1993

1993
[14]

L. Lamport. LaTeX: A Document Preparation System. 1994

1994
[15]

Metropolis and A

N. Metropolis and A. W. Rosenbluth and M. N. Rosenbluth and A. H. Teller and E. Teller. Equations of state calculations by fast computing machine. J. Chem. Phys. 1953

1953
[16]

L. C. Perelman and J. Paradis and E. Barrett. Mayfield Handbook of Technical and Scientific Writing. 1997

1997
[17]

John D Lees-Miller , title =

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [1]

, keywords =

Direction-dependent calibration with image-domain gridding. , keywords =. doi:10.1051/0004-6361/202558272 , archivePrefix =. 2602.06002 , primaryClass =

work page arXiv

[2] [2]

Ionospheric calibration of low frequency radio interferometric observations using the peeling scheme. I. Method description and first results. , keywords =. doi:10.1051/0004-6361/200811094 , archivePrefix =. 0904.3975 , primaryClass =

arXiv

[3] [3]

Radio Science , keywords =

Scintillations due to a concentrated layer with a power-law turbulence spectrum. Radio Science , keywords =. doi:10.1029/RS010i001p00107 , adsurl =

work page doi:10.1029/rs010i001p00107

[4] [4]

Correcting direction-dependent gains in the deconvolution of radio interferometric images

Correcting direction-dependent gains in the deconvolution of radio interferometric images. , keywords =. doi:10.1051/0004-6361:20079284 , archivePrefix =. 0805.0834 , primaryClass =

work page Pith review arXiv

[5] [5]

, keywords =

LOFAR Facet Calibration. , keywords =. doi:10.3847/0067-0049/223/1/2 , archivePrefix =. 1601.05422 , primaryClass =

arXiv

[6] [6]

Entropy Encoding, Hilbert Space and Karhunen-Loeve Transforms

Entropy encoding, Hilbert space, and Karhunen-Lo \`e ve transforms. Journal of Mathematical Physics , keywords =. doi:10.1063/1.2793569 , archivePrefix =. math-ph/0701056 , primaryClass =

work page Pith review arXiv

[7] [7]

, TITLE=

Guio, Patrick and Pécseli, Hans L. , TITLE=. Frontiers in Astronomy and Space Sciences , VOLUME=. 2021 , URL=. doi:10.3389/fspas.2020.573746 , ISSN=

work page arXiv 2021

[8] [8]

Ionospheric Calibration for the LOFAR Radio Telescope , year=

van der Tol, Sebastiaan and van der Veen, Alle-Jan , booktitle=. Ionospheric Calibration for the LOFAR Radio Telescope , year=

work page