{"id":"6cd0e91f-ed77-4883-8ce6-716fd815403b","arxiv_id":"2608.10762","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"The paper derives a more efficient computation order and statistically optimized basis functions for IDG-CAL, without implementing or testing them.","lead":"This paper suggests two ways to speed up and improve a radio calibration method: a smarter calculation order and statistically designed correction shapes. It shows the math but does not yet test the ideas on real data.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'optimal' KL basis in §4.2 rests on an admitted stand-in: Eq. (9) assumes a Kolmogorov structure function with unspecified β and is stated in §4.2 to be inaccurate for instrumental effects, with no empirical fit or sensitivity test; if real gain fluctuations deviate, the second improvement is…","rationale":"The reader's weakest assumption identified the Kolmogorov structure function in Eq. (9) as the main unverified input to the KL basis. My reading agrees: the paper explicitly disclaims the model for instrumental effects, leaving the second improvement's optimality unsupported. I do not see a reason to move the verdict from CONDITIONAL; the concern reinforces the need for revision rather than outright rejection, because the paper is a methods proposal and the stochastic-model substitution is acknowledged. The Eq. (11) typo is an additional reproducibility problem but not a separate conceptual objection. The concrete test would settle whether the model mismatch is load-bearing by comparing the proposed basis with an empirically derived basis on the actual dataset. Since the reader already made the conditional verdict and my concern does not change that disposition, the verdict is unchanged.","tokens_in":5001,"tokens_out":25770,"duration_ms":269624,"concrete_test":"Fit the empirical spatial structure function of the estimated gain screens from a high-order calibration on the real 8-hour dataset; compare it to Eq. (9) for β in [0.5,2]. Then repeat the imaging/calibration with KL bases built from (a) Eq. (9) for several β, (b) the empirical covariance, and (c) polynomials, and compare background noise and artifacts. If the Eq. (9)-based basis is measurably worse than the empirical basis, the optimality claim is unsupported. Also recompute the basis with the corrected distance norm to isolate the Eq. (11) typo.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's second claimed improvement is that Eqs. (11)-(15) produce an optimal set of basis functions for the A-terms. That optimality is conditional on the stochastic model in Eq. (9), a power-law structure function with exponent β. Section 4.2 explicitly concedes that for instrument-based effects this model is 'not an accurate description of the physics', and the paper gives no procedure for choosing β or for testing the model against data. Since the motivating full-dataset result is limited by real instrumental and ionospheric fluctuations, the derived KL functions are not demonstrated to be optimal for the actual gain screen. The text says only that the basis 'might still be good enough', which is an admitted assumption, not a result. Compounding this, Eq. (11) as printed uses ((l1-l2)^2(m1-m2)^2)^{β/2} rather than ((l1-l2)^2+(m1-m2)^2)^{β/2}, so the matrix is not the claimed Kolmogorov structure function; as written the derivation is not reproducible. Because the paper contains no implementation or test of either improvement, the load-bearing condition for the second improvement is an unverified stochastic model plus a typo in the key matrix.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","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.","tokens_in":5285,"tokens_out":3554,"duration_ms":39097,"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":[{"comment":"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.","section":"§4.2, Eq. (11)"},{"comment":"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.","section":"§4.2, Eqs. (9)-(15)"},{"comment":"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.","section":"§4.1 and §5"}],"minor_comments":[{"comment":"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.","section":"§1 and throughout"},{"comment":"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.","section":"§2, Eq. (1)"},{"comment":"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.","section":"§4.1, Eq. (3)"},{"comment":"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.","section":"§4.2, Eqs. (12)-(15)"},{"comment":"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).","section":"§4.2, Eq. (9)"},{"comment":"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.'","section":"§5"}],"recommendation":"major_revision","confidential_remarks":"This manuscript reads more like a status report or proceedings contribution than a completed research paper. The mathematical derivation in Section 4.1 appears sound, and the proposed KL basis approach is a legitimate idea, but the lack of any implementation or validation is a serious gap for a journal submission. Major revision is warranted if the authors can fix the typo in Eq. (11), provide at least a synthetic-data or small-scale test of both improvements, and discuss the choice of β. If no such tests can be added, the paper would be more appropriate for a workshop/proceedings venue than for a fully refereed journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis 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.\n\nThe 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.\n\nThe 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.\n\nWho 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.\n\nRecommendation: accept for peer review, with expected heavy revision.","headline":"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.","tokens_in":5777,"tokens_out":2704,"would_cite":false,"duration_ms":26289,"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 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.","keywords":["radio interferometry","wide-field imaging","direction-dependent calibration","image-domain gridding","A-projection","Karhunen-Loève basis functions","IDG-CAL"],"falsifier":"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.","tokens_in":4810,"feed_emoji":"📡","tokens_out":7203,"duration_ms":58237,"temperature":0.7,"pith_summary":"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.","feed_headline":"Two fixes make IDG-CAL scale to deeper wide-field images","feed_subtitle":"Reordering the parameter loop and switching to turbulence-based basis functions cut cost and improve A-term detail.","key_machinery":"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.","core_discovery":"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.","pith_inferences":["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."],"forward_implications":["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."],"supporting_citations":[{"why":"supplies the IDG-CAL algorithm, the cost function derivative in its Eq. (7)/(14), and the previous small-dataset results that this paper aims to scale up.","marker":"[1]"},{"why":"provides the facet-calibration baseline (50 facets) against which IDG-CAL's background and artifacts are compared.","marker":"[2]"},{"why":"establishes the A-projection gridding-kernel approach that IDG-CAL's continuous screen corrections extend.","marker":"[3]"},{"why":"supplies the Karhunen-Loève transform theory used to derive the optimal basis functions.","marker":"[4]"},{"why":"motivates the Kolmogorov-turbulence structure function used to build the correlation matrix.","marker":"[5]"}],"fun_headline_variants":["Proposed IDG-CAL tweaks cut cost, add detail","IDG-CAL: cheaper deep imaging via two fixes","Two proposed changes boost IDG-CAL for deep fields","IDG-CAL deep imaging: reorder loops, new basis"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"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.","fun_headline_variants_meta":{"raw":{"variants":["Proposed IDG-CAL tweaks cut cost, add detail","IDG-CAL: cheaper deep imaging via two fixes","Two proposed changes boost IDG-CAL for deep fields","IDG-CAL deep imaging: reorder loops, new basis"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000229,"raw_usage":{"total_tokens":1433,"prompt_tokens":856,"completion_tokens":577,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":472,"completion_tokens_details":{"reasoning_tokens":506}},"tokens_in":472,"tokens_out":577,"duration_ms":6408,"temperature":1.0,"reasoning_tokens":506,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T17:53:15.264518+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"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.","supporting_citations":[{"cited_title":", keywords =","cited_arxiv_id":null,"evidence_quote":"supplies the IDG-CAL algorithm, the cost function derivative in its Eq. (7)/(14), and the previous small-dataset results that this paper aims to scale up."},{"cited_title":"Radio Science , keywords =","cited_arxiv_id":null,"evidence_quote":"establishes the A-projection gridding-kernel approach that IDG-CAL's continuous screen corrections extend."},{"cited_title":"Correcting direction-dependent gains in the deconvolution of radio interferometric images","cited_arxiv_id":"0805.0834","evidence_quote":"supplies the Karhunen-Loève transform theory used to derive the optimal basis functions."}],"review_version":1}