{"id":"7969fcc2-1e63-411e-9a78-00b7b1ddb987","arxiv_id":"1908.02516","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Using the largest observed earthquake, the paper shows the corner magnitude of the global seismic-moment distribution can be constrained within about 80 years, not the 200,000 years claimed in 2013.","lead":"This paper estimates how many more years of earthquake recording are needed to pin down the size of the largest possible earthquakes. It finds that, contrary to a 2013 claim, the end of this century should be enough to narrow the range of the corner magnitude, provided current models and rates hold.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Table 1's 2097 rows are conditional scenarios, but the abstract/conclusion says the range 'will' shrink. Under the Tap model at the current upper bound mc=10.2, P(M≥10 by 2097)≈56%; a 10.0 maximum would leave a ~9.7–∞ range, so the headline overstates what is demonstrated.","rationale":"The paper makes a valid correction: inverting Eq. (5) via Eqs. (10)-(11) is a coherent way to turn the maximum statistic into compatible ranges, and Fig. 2 correctly shows that Zöller's decreasing data requirement is an artifact of Eq. (8). The reader's independence concern is real but secondary: even a factor-two reduction in effective N leaves the 2097 ranges bounded for the scenarios shown, so it qualifies the numbers without overturning the direction. The random-future-maximum issue is more load-bearing because it threatens the headline 'will be reduced.' The analysis itself is explicit about conditionality ('will strongly depend on the maximum magnitude observed'), but the abstract and conclusions do not keep that qualification, and no probability calculation is offered. Thus the appropriate verdict remains CONDITIONAL: the statistical machinery is sound, but the headline needs softening or the missing probability calculation must be supplied.","tokens_in":14950,"tokens_out":25726,"duration_ms":266756,"concrete_test":"Recompute the 2097 rows of Table 1 for hypothetical maxima m=9.7 and m=10.0 with N=25,644 and β=0.67. Then, for each model and for each mc in the current 95% compatible range, compute P(M_max(2097) ≥ m | mc) using the Poisson occurrence model; report the probability that the resulting compatible set is unbounded above (e.g., by checking whether the upper endpoint is ∞ for m=10.0). If that probability is non-negligible, replace the unconditional future statement with an explicitly conditional one.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing gap is the step from Table 1's conditional scenarios to the abstract's assertion that 'the range will be substantially reduced by the end of this century.' Table 1 lists compatible mc ranges only for hypothetical future maxima of 9.1, 9.3, and 9.5; it does not assign probabilities to those maxima. Under models still compatible with current data, the unconditional claim can fail. For example, the Tap model with mc=10.2 (the upper end of the 2017 compatible range) implies S(10)≈3.2×10^-5, hence an expected ~0.83 events of magnitude ≥10 in the 120 years from 1977 to 2097, i.e., a probability >50% of at least one such event. If a magnitude 10.0 maximum is observed by 2097, the Tap compatible range shifts to roughly mc≈9.7–∞ (and similarly TPL to 10–∞), so the range is not substantially reduced. Because the paper does not integrate over the Poisson-distributed future maximum or compute the probability of an unbounded range, the 'will' claim is stronger than the analysis supports.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper revisits the question of how much additional earthquake recording is needed to constrain the corner value (Mc or mc) of the global seismic-moment distribution. Using the global CMT catalog (7,585 events with magnitude ≥5.75, 1977–2012.5, extended to 2017) and fixing β=0.67, the authors compute 95% probability intervals for the maximum of N independent observations under three models — truncated power law (TPL), tapered GR (Tap), and truncated gamma (TrG) — via Fmax(y)=[F(y)]^N. From this they derive currently compatible ranges of mc (e.g., 8.6–10.2 for Tap) and conditional future ranges for hypothetical maxima in 2047 and 2097, and they argue that these ranges will shrink substantially by 2097, contrary to earlier claims that hundreds of thousands of years would be needed. They also show that the 'well-sampledness' criterion used by Zöller (2013) is statistically flawed.","tokens_in":15227,"tokens_out":5796,"duration_ms":56482,"significance":"The paper makes a useful methodological correction: the earlier 'well-sampledness' criterion (Zöller 2013) is replaced by a proper extremal significance test based on the distribution of the maximum, and the forward calculation is not circular because mc is not fitted to the observed maximum. The central probability calculation under the stated independence assumption is standard and appears correct, and the resulting Table 1 gives concrete, falsifiable conditional predictions. However, the headline claim that the range 'will' shrink by 2097 is not established by the conditional analysis.","major_comments":[{"comment":"The abstract and Conclusions assert that 'under reasonable assumptions, the range will be substantially reduced by the end of this century,' but Table 1 only lists compatible mc ranges conditional on hypothetical future maxima of 9.1, 9.3, and 9.5; no probabilities are assigned to these maxima. This is load-bearing, because models still compatible with current data can produce a future maximum that breaks the claimed reduction. For instance, the Tap model with mc=10.2 (the upper end of the 2017 compatible range) has S(10)≈3.2×10^-5, giving an expected 0.83 events of magnitude ≥10 in 120 years and a probability >50% of at least one; under a 10.0 maximum, the Tap compatible range shifts to roughly mc≈9.7–∞, so the range is not substantially reduced. The unconditional claim therefore needs either to be weakened to a conditional one or derived by integrating over the Poisson-distributed future maximum using a prior over mc.","section":"Abstract/Conclusions; Table 1 in 'Proper constraining of the corner seismic-moment: Tap and TrG cases'"},{"comment":"Eq. (5) and all subsequent percentile calculations assume the 7,585 events are independent. The paper acknowledges this once ('assuming ... there is no dependence between the magnitudes') but does not quantify the effect of aftershock clustering on the 2047/2097 ranges. Since N enters Eqs. (10)–(11) and Table 1 through p^{1/N}, a modest reduction in the effective number of independent events can materially widen the reported intervals; a sensitivity analysis (e.g., using a declustered catalog or an effective N) is needed before the projections can be considered robust.","section":"Eq. (5); 'Proper constraining of the corner seismic-moment: TPL case'"},{"comment":"All entries use β=0.67, stated to be 'very close to the maximum-likelihood solution,' but no uncertainty in β is propagated. The compatible ranges of mc are monotone in β through the percentile formulas (10)–(11), so an error of, say, 0.05 in β could shift the bounds by several tenths of a magnitude. The paper needs at least a sensitivity analysis for β or a justification for treating it as known.","section":"Throughout; Figs. 2–3 and Table 1"}],"minor_comments":[{"comment":"Eq. (12) writes Stpl(x|Mc) in the integrand while the surrounding text is about the tapered model; this should be Stap (or a generic S_model).","section":"Discussion, Eq. (12)"},{"comment":"The word 'assessment' is misspelled as 'assesment'.","section":"Discussion, first sentence"},{"comment":"The caption uses the notation 'mpm' and '(mp, mp+0.95)' without defining which interval is symmetric and which is of minimum width; a brief explanation would improve readability.","section":"Figure 3 caption"},{"comment":"The captions misspell 'corresponding' as 'corresponing'.","section":"SI captions, Figs. S4–S6"},{"comment":"The sentence 'we need to wait about 30 years to chose between these three answers' contains a typo: 'to chose' should be 'to choose'.","section":"Section 'Proper constraining of the corner seismic-moment: TPL case', last paragraph"}],"recommendation":"major_revision","confidential_remarks":"The correction of Zöller's criterion is a genuine contribution and the forward calculation is transparent, so the paper is within the journal's scope. The main risk is the overstatement of the central claim; if the authors revise the abstract/conclusions to be conditional, or properly derive an unconditional statement, the paper would be acceptable. There are no citation or novelty concerns."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this paper fixes a real error in Zöller's 2013 estimate and gives a much more sensible way to think about how long we need to wait to pin down the corner magnitude. The core move—replacing Zöller's ad-hoc \"well-sampledness\" condition with proper 95% intervals from the distribution of the maximum—is correct and overdue. For the TPL model, the maximum is the MLE of the truncation parameter, so the percentile inversion gives genuine confidence intervals for mc. The numerical contrast is stark: Zöller's 200,000 years becomes on the order of 65–80 years for plausible corner values. Table 1 is a clearly presented conditional forecast of what different future maxima would do to the compatible range. That is real, citable value for statistical seismology.\n\nThe soft spots are real but not fatal. The main one is the abstract's claim that the range \"will be substantially reduced by the end of this century.\" Table 1 only shows scenarios with future maxima of 9.1, 9.3, or 9.5. It does not integrate over the distribution of the future maximum. The stress-test calculation is correct: the Tap model with mc=10.2 is still compatible with current data, and under that model P(M≥10 by 2097) is above 50%. If a magnitude 10 event occurs, the compatible Tap range shifts to roughly 9.6–∞, which is not a substantial reduction—the upper bound stays unbounded. So the unconditional statement in the abstract overstates what the analysis demonstrates. The conclusions are more careful, saying \"expected to decrease substantially, but depending crucially on the maximum,\" but the abstract needs softening or the claim needs to be phrased as conditional on the future maximum being in a specific range.\n\nSecond, Eq. (5) assumes independence of the 7,585 magnitudes. The paper acknowledges this in one sentence but does not test sensitivity to clustering. Aftershock sequences reduce the effective sample size and widen the intervals; this is a moderate concern that should be addressed with a declustered catalog or simulation before the numbers are taken as operational. Third, β is fixed at 0.67 with no uncertainty propagation. Since β enters every formula, its estimation error should widen the reported ranges. Minor, but worth a sentence.\n\nOverall: the central statistical argument holds up; the correction to Zöller is legitimate; the forward calculation is not circular. The paper deserves serious peer review, but the abstract and conclusions should be revised to match the conditional nature of Table 1, and a sensitivity analysis for clustering and β would strengthen it. I would bring this to a reading group and would cite it in my own work as the corrected reference on time-to-constraint for corner parameters.","headline":"Useful statistical correction to Zöller's 200,000-year estimate, but the abstract's 'will shrink by 2097' only holds if the future maximum lands in a narrow window.","tokens_in":15730,"tokens_out":3903,"would_cite":true,"duration_ms":39013,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A few more decades of global earthquake records can tightly constrain the upper tail of the quake-size distribution.","keywords":["seismic moment distribution","corner magnitude","Gutenberg-Richter law","extreme value statistics","maximum order statistic","truncated power law","tapered distribution","earthquake catalogs"],"falsifier":"Recompute the Table 1 ranges after removing aftershocks from the global CMT catalog; if declustering moves the 95% intervals by more than a few tenths of a magnitude or pushes the empirical maximum outside them, the independence assumption is the weak link. The long-run check is to record the actual maximum to 2097 and see which of the 9.1, 9.3, or 9.5 branches of Table 1 it selects.","tokens_in":14753,"feed_emoji":"🌍","tokens_out":11902,"duration_ms":112686,"temperature":0.7,"pith_summary":"This paper asks how many more years of global earthquake recording are needed before the tail of the earthquake-size distribution can be inferred. It argues that the answer is decades, not hundreds of thousands of years: the range of allowed values of the corner magnitude, the scale at which the Gutenberg-Richter power law bends downward, will narrow substantially by the end of this century once the next very large earthquake is observed. The argument uses the largest event in the global CMT catalog as the test statistic and compares it with the predicted distribution of the maximum of N earthquakes for three tail models: the truncated power law, the tapered Gutenberg-Richter law, and the truncated gamma. As of 2017 the compatible range of the corner magnitude is wide, but Table 1 projects, for example, that if the largest event by 2097 has magnitude 9.3, the tapered model's corner magnitude will be constrained to about 8.8-9.4. Because hazard assessments depend on how fast the tail decays, knowing when this parameter can be resolved affects the credibility of long-term seismic hazard estimates.","feed_headline":"Decades, not millennia, to pin down the biggest possible earthquake","feed_subtitle":"The biggest quake seen by 2097 will shrink the uncertainty on the true maximum quake size from unbounded to a narrow band.","key_machinery":"The load-bearing object is the maximum-order-statistic identity Fmax(y)=Prob[Y <= y]=[F(y)]^N, the cumulative distribution of the largest of N independent seismic moments. For the truncated power-law model, the maximum Y is exactly the maximum-likelihood estimator of the truncation parameter, so inverting Fmax(yp)=p gives 95% probability intervals for the estimated corner. For the tapered and truncated-gamma models the same percentile inversion is performed through closed forms involving implicit functions (the Lambert W function for the tapered model and the inverse incomplete gamma for the truncated gamma), and the empirical maximum is compared with those intervals to decide which corner values are compatible with the data. The machinery also supplies the sample size N needed to shrink an interval to a chosen width by solving the width equation; this replaces the earlier 'well-sampledness' criterion, which the paper shows is not a valid statistical test because it treats the observed maximum as fixed while N grows.","core_discovery":"Contrary to an earlier analysis of the same catalog that concluded reliable estimation of the tapered corner magnitude would require about 200,000 years, the authors show that the correct use of the largest observed earthquake gives a much shorter horizon. The distribution of the maximum of N independent events is Fmax(y)=[F(y)]^N, so the observed largest magnitude can be tested against percentile intervals for each candidate model and each value of the corner magnitude. Inverting those percentiles yields, for each hypothetical true corner, the number N (or calendar year) at which the 95% interval narrows to a chosen width. Applying this to the 7,585 CMT events since 1977 leaves the corner magnitude unbounded above for all three models at the original catalog length, but already gives finite ranges by 2017, and under the paper's stated assumptions the projected 2097 ranges are much narrower: for the tapered model they are between about 0.5 and 1.0 magnitude wide, and for the truncated gamma between 0.5 and 1.4. The paper's Table 1 gives the predicted ranges for 2047 and 2097 conditional on the maximum magnitude observed in the interval (9.1, 9.3, or 9.5), showing that the outcome depends strongly on whether an event of magnitude 9.3 or larger occurs.","pith_inferences":["If aftershock clustering is strong enough to reduce the effective number of independent events below 7,585, the predicted narrowing dates would shift later than 2047 or 2097; this is testable because declustered catalogs would give wider percentile intervals.","The same maximum-based percentile logic transfers to any heavy-tailed record where the scale parameter is debated and only a handful of extremes exist, such as storm surges, floods, or geomagnetic disturbances, provided a candidate tail family is specified.","The calendar projections assume a constant global rate of 213.7 events per year and a fixed beta; a persistent rate change would move the dates but would not change the N-based widths, so the method is more robust than the specific years printed in Table 1.","The mixture over corner values in Eq. (12) is equivalent to a Bayesian posterior under a 1/M_c prior; making that prior explicit would let hazard calculators report full predictive distributions instead of range endpoints."],"forward_implications":["By the end of 2017 the CMT data already give finite 95% ranges for the corner magnitude of the tapered (8.6-10.2), truncated power law (9.1-10.8), and truncated gamma (8.8-11.2) models, whereas the same data at mid-2012 allowed no upper bound.","If no earthquake larger than magnitude 9.1 occurs before 2047, the truncated power law narrows the corner to 9.1-9.5; if a 9.3 event occurs instead, the tapered model allows 8.8-9.95, so the observation of a very large event is the main information carrier.","The earlier conclusion that hundreds of thousands of years are needed for the tapered model is reversed; under the paper's assumptions, the 2097 ranges are narrow enough to distinguish among competing hypotheses about the tail.","Seismic hazard should be evaluated by mixing tail models over the range of compatible corner values (Eq. 12) rather than by relying on a point estimate or on asymptotic standard errors, which the paper argues misdescribe the uncertainty."],"supporting_citations":[{"why":"The previous study whose 'well-sampledness' criterion and 200,000-year estimate this paper re-examines and overturns.","marker":"[16]"},{"why":"Proposed a tapered-model corner magnitude near 9, the value the earlier study rejected and the present analysis finds compatible.","marker":"[14]"},{"why":"Previous fits of the global seismic-moment distribution whose asymptotic standard errors the paper argues do not describe the uncertainty in the corner parameter.","marker":"[20]"},{"why":"Supplies the global CMT catalog, the 7,585 events, and the empirical maximum magnitude 9.1 used throughout.","marker":"[25]"},{"why":"Introduces the truncated-gamma model used as one of the three candidate tail distributions.","marker":"[12]"},{"why":"Defines the corner seismic moment and motivates why the Gutenberg-Richter tail must bend.","marker":"[9]"},{"why":"Documents the 1960 Chile event of about magnitude 9.5, used to argue that a no-event-above-9.1 scenario to 2097 is implausible for the truncated power law.","marker":"[28]"},{"why":"Provides the implicit function used to invert the tapered-model maximum percentiles.","marker":"[29]"},{"why":"Explains the multiple-source magnitude 9.3 for the 2004 Sumatra earthquake, justifying the catalog's 9.1 observed maximum.","marker":"[27]"},{"why":"Textbook source for the distribution of the maximum of N independent observations, the foundation of Eq. (5).","marker":"[22]"}],"fun_headline_variants":["Quake size limit narrowed by 2097, not 200,000 years","Corner magnitude constrained within decades, not millennia","Biggest quake uncertainty shrinks by century's end","Rethinking quake tail: corner value pinned sooner","Earthquake maximum clearer by 2097"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The projected time windows assume the 7,585 catalogued earthquake magnitudes are independent, so the chance that the largest observed event stays below a value is the Nth power of the single-event chance; global catalogs contain aftershock clustering, and the paper states but does not test this dependence assumption.","fun_headline_variants_meta":{"raw":{"variants":["Quake size limit narrowed by 2097, not 200,000 years","Corner magnitude constrained within decades, not millennia","Biggest quake uncertainty shrinks by century's end","Rethinking quake tail: corner value pinned sooner","Earthquake maximum clearer by 2097"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000556,"raw_usage":{"total_tokens":2665,"prompt_tokens":984,"completion_tokens":1681,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":600,"completion_tokens_details":{"reasoning_tokens":1601}},"tokens_in":600,"tokens_out":1681,"duration_ms":13483,"temperature":1.0,"reasoning_tokens":1601,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T14:41:48.288221+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute the Table 1 ranges after removing aftershocks from the global CMT catalog; if declustering moves the 95% intervals by more than a few tenths of a magnitude or pushes the empirical maximum outside them, the independence assumption is the weak link. The long-run check is to record the actual maximum to 2097 and see which of the 9.1, 9.3, or 9.5 branches of Table 1 it selects.","supporting_citations":[{"cited_title":"Convergence of the frequency-magnitude distribution of global earthquakes: Maybe in 200 years","cited_arxiv_id":null,"evidence_quote":"The previous study whose 'well-sampledness' criterion and 200,000-year estimate this paper re-examines and overturns."},{"cited_title":"Convergence of the frequency-size distribution of global earthquakes","cited_arxiv_id":null,"evidence_quote":"Proposed a tapered-model corner magnitude near 9, the value the earlier study rejected and the present analysis finds compatible."},{"cited_title":"Deviation from power law of the global seismic moment distribution","cited_arxiv_id":null,"evidence_quote":"Previous fits of the global seismic-moment distribution whose asymptotic standard errors the paper argues do not describe the uncertainty in the corner parameter."},{"cited_title":"The global CMT project 2004-2010: Centroid-moment tensors for 13,017 earthquakes","cited_arxiv_id":null,"evidence_quote":"Supplies the global CMT catalog, the 7,585 events, and the empirical maximum magnitude 9.1 used throughout."},{"cited_title":"Effect of the Sumatran mega-earthquake on the global magnitude cut-off and event rate","cited_arxiv_id":null,"evidence_quote":"Introduces the truncated-gamma model used as one of the three candidate tail distributions."},{"cited_title":"Seismic moment distribution revisited: I","cited_arxiv_id":null,"evidence_quote":"Defines the corner seismic moment and motivates why the Gutenberg-Richter tail must bend."},{"cited_title":"Long-Term Perspectives on Giant Earthquakes and Tsunamis at Subduction Zones","cited_arxiv_id":null,"evidence_quote":"Documents the 1960 Chile event of about magnitude 9.5, used to argue that a no-event-above-9.1 scenario to 2097 is implausible for the truncated power law."},{"cited_title":"On the Lambert W function","cited_arxiv_id":null,"evidence_quote":"Provides the implicit function used to invert the tapered-model maximum percentiles."},{"cited_title":"Multiple CMT source analysis of the 2004 Sumatra earthquake","cited_arxiv_id":null,"evidence_quote":"Explains the multiple-source magnitude 9.3 for the 2004 Sumatra earthquake, justifying the catalog's 9.1 observed maximum."},{"cited_title":"A First Course in Probability","cited_arxiv_id":null,"evidence_quote":"Textbook source for the distribution of the maximum of N independent observations, the foundation of Eq. (5)."}],"review_version":1}