pith. sign in

arxiv: 2605.30492 · v1 · pith:FRZWW6Z7new · submitted 2026-05-28 · 📊 stat.ME · stat.CO

Shrinkage-Constrained Functional Calibration for Complex Computer Models

Liam Myhill , Enrique Martinez , Sez Russcher This is my paper

Pith reviewed 2026-06-29 05:40 UTC · model grok-4.3

classification 📊 stat.ME stat.CO
keywords Bayesian model calibrationGaussian processshrinkage priorsKennedy-O'Hagan frameworkmodel discrepancyfunctional calibrationorthogonality constraintsinput-dependent parameters
0
0 comments X

The pith

A new calibration formalism represents each parameter as a fixed best estimate plus an input-dependent correction GP under strong shrinkage priors, nesting KOH while extending it to functional calibration.

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

The paper develops integrated bias with full uncertainty as a Bayesian alternative to the Kennedy-O'Hagan framework for calibrating complex computer models. Calibration parameters are expressed as the sum of a known best estimate and a correction term given by an independent Gaussian process over the input domain, equipped with conservative hyperpriors that enforce shrinkage. An additive discrepancy Gaussian process captures any remaining mismatch, and orthogonality constraints are imposed to limit confounding between the simulator, the correction, and the discrepancy. When the shrinkage is tight the correction stays flat and the method recovers KOH predictions; when the data support variation the correction is permitted to become a smooth function of the inputs in a controlled fashion. A reader would care because the construction supplies stronger regularization and reduces the confounding pathologies that arise with sparse noisy observations of inexact simulators.

Core claim

The authors argue that anchoring each calibration parameter around a best estimate via a shrinkage-constrained input-dependent Gaussian process correction, together with an additive discrepancy process and orthogonality constraints, produces a model that nests the standard KOH formulation as the special case of strong shrinkage while extending it to input-dependent calibration whenever the data justify relaxing the shrinkage.

What carries the argument

The parameter correction Gaussian process over the input space, equipped with conservative hyperpriors on complexity and combined with orthogonality constraints between simulator, correction, and additive discrepancy.

If this is right

  • Under strong shrinkage the posterior predictions converge to those of the KOH model.
  • When shrinkage is relaxed the mean correction becomes an explicit function of the inputs if the data support it.
  • The formulation supplies active regularization around the supplied best estimates rather than weak priors on fixed parameters.
  • Orthogonality constraints reduce collinearity between the simulator output, the parameter correction, and the additive discrepancy.
  • The same structure applies to any computer model for which a best estimate of each calibration parameter is available.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same shrinkage construction could be applied to other discrepancy terms beyond calibration parameters, such as input-dependent model error in surrogate modeling.
  • If the method recovers known input-dependent effects on synthetic data, it would support use in engineering domains where physical parameters vary with operating conditions.
  • The orthogonality constraints might be relaxed or replaced by other identifiability penalties without losing the nesting property.

Load-bearing premise

An independent Gaussian process with conservative hyperpriors can represent any needed input-dependent correction to the calibration parameters while the orthogonality constraints keep confounding between the simulator, correction, and additive discrepancy under control.

What would settle it

A controlled simulation in which the true calibration parameter is known to vary smoothly with an input variable; the method would be falsified if the posterior mean of the correction term fails to recover that variation or if the additive discrepancy term absorbs most of the signal.

Figures

Figures reproduced from arXiv: 2605.30492 by Enrique Martinez, Liam Myhill, Sez Russcher.

Figure 1
Figure 1. Figure 1: Datasets for Studies No. 1 and 2 True Problem Parameters Utilized Shrinkage Hyperparameters Study No. 𝜽𝟏 (𝒙) 𝜽𝟐 (𝒙) 𝜹𝜼 (𝒙) 𝓵𝜿 𝝈𝜿 𝓵𝜹 𝝈𝜹 1 1.5 1.75 cos(𝑥) 10 0.1 0.1 0.1 2 sin(𝑥) cos(𝑥) 0.5 0.2 0.1 0.1 MD–DDD – – – 10 0.1 0.1 0.1 Hyperprior Distributions: 𝓁𝜅 , 𝓁𝛿 ∼ (𝜇 = 0.3, 𝜎 = 0.5), 𝜎2 𝜅 , 𝜎2 𝛿 ∼ (𝜇 = 0.05, 𝜎 = 0.5) [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: IBFU posterior predictions of Study No. 1 5. Calibration of Continuum Crystal Plasticity Models A real-world application is also presented, whereby IBFU is used to calibrate mesoscale defect models to atomistic simulation data, which is treated as experimental observation. Molecular Dynamics (MD) simulations feature simulation cells containing millions of degrees of freedom for length-scales in the range o… view at source ↗
Figure 3
Figure 3. Figure 3: Integrated bias results of Study No. 1 (a) KOH posterior predictions of Study No. 1 1 0 10 20 30 40 50 60 1 2 0.5 1 1.5 0.5 1 1.5 2 0.0025 0.4981 0.9938 1.4895 1.9851 Posterior Calibration Parameter Density (b) Posterior parameter distribution output by KOH, where stars and red dashed lines denote true values [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: KOH results of Study No. 1 calibrated to capture the same trends observed in MD at the microscale, and extrapolate such behaviors to the meso￾macroscale. When studying the plastic deformation of crystalline solids, Discrete Dislocation Dynamics (DDD) is the natural candidate model to facilitate the projection of crystal defect behavior to larger scales. In both MD and DDD, the critical resolved shear stres… view at source ↗
Figure 5
Figure 5. Figure 5: Results of IBFU applied to Study No. 2 terms {𝜂(𝑥, 𝜃0 ), 𝜂(𝑥, 𝜃0+𝜅(𝑥)), 𝛿𝜂 (𝑥)}. The results of the analysis are given in [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Integrated bias results of Study No. 2 (a) KOH posterior predictions of Study No. 2 3 1 3 1 0 10 20 30 40 50 60 70 3 2 0.5 1 1.5 3 2 0.5 1 1.5 0.0015 0.4999 0.9983 1.4967 1.9951 Posterior Calibration Parameter Density (b) Posterior parameter distribution output by KOH with high￾lighted true region [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: KOH results of Study No. 2 Application of IBFU to a practical calibration problem faced by the material science community demonstrates its ability to differentiate the various sources of error introduced when coarse-graining the multiplicity of physical mechanisms present at the atomistic scale. 6. Discussion One significant challenge faced during the application of IBFU to the demonstrated conceptual prob… view at source ↗
Figure 8
Figure 8. Figure 8: IBFU predictions of the various error sources in the calibration of continuum DDD model predictions of 𝜏 𝐶𝑅𝑆𝑆 to MD observations Myhill et al. [2026] no-curvature, moderate-amplitude response necessitated for each 𝜅 discrepancy field, 𝓁𝜅 ought to be relatively large (𝑂([10 − 100]), and 𝜎𝜅 kept to 𝑂(0.1) to remain proportional with the degree of confidence in the magnitude of model predictions. Supplementin… view at source ↗
Figure 9
Figure 9. Figure 9: Relative contributions of 𝜅 and 𝛿𝜂 GPs in generating posterior predictions of 𝜏 𝐶𝑅𝑆𝑆 should take the form of a complexity shrinkage on 𝜅(𝑥) (increase in 𝓁𝜅 ). Such constraints should only be relaxed if one can argue that parameter drift is a physically meaningful mechanism for addressing discrepancies. If it is known a-priori that model predictions are far from experimental observations, increasing 𝜎𝛿 will… view at source ↗
Figure 10
Figure 10. Figure 10: Sensitivity analysis of internal GP models to hyperparameters {𝜙𝜅 , 𝜙𝛿𝜂 }. The left column corresponds to Study No. 1 and the right column corresponds to Study No. 2. : Preprint submitted to Elsevier Page 18 of 23 [PITH_FULL_IMAGE:figures/full_fig_p018_10.png] view at source ↗
read the original abstract

We propose a new Bayesian model calibration formalism as an alternative to the Kennedy O'Hagan (KOH) framework which we term integrated bias with full uncertainty (IBFU). In KOH, calibration parameters are modeled as fixed, but unknown distributions with relatively weak prior constraints, and their posteriors are inferred jointly with an additive discrepancy Gaussian Process (GP). This formulation often provides limited regularization and leads to confounding pathologies when applied to inexact models with sparse, noisy measurements. By contrast, we represent each calibration parameter as the sum of a fixed best estimate value and a parameter correction represented by an independent GP over the input space, equipped with strong shrinkage priors. Any residual discrepancy that cannot be addressed via parameter correction is captured by an additive discrepancy GP operating on the simulator, similar to KOH. We then impose orthogonality constraints to mitigate confounding between the simulator and modeled additive discrepancy and colinearity between model parameters. Imposing strong complexity shrinkage via conservative hyperpriors forces the mean parameter correction to remain flat across the domain, resulting in predictions that essentially converge with the KOH formulation. However, upon relaxing complexity shrinkage, should the data provide evidence that the effective calibration parameter varies across the domain, the mean parameter correction is allowed to become a function of the domain in a controlled, structured manner. In this sense, our approach is more universal: it effectively nests KOH as a special case while extending it to input dependent calibration, and it is more tightly constrained, because it anchors the true values around the best estimates and the shrinkage prior actively regularizes the calibration parameters.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes the Integrated Bias with Full Uncertainty (IBFU) formalism as an alternative to the Kennedy-O'Hagan (KOH) framework for Bayesian calibration of complex computer models. It models each calibration parameter as a fixed best estimate plus an input-dependent correction represented by an independent GP equipped with strong shrinkage priors; residual discrepancy is captured by an additive discrepancy GP, with orthogonality constraints imposed to mitigate confounding between the simulator output, the parameter correction, and the discrepancy term. The approach is claimed to nest KOH as a special case under strong shrinkage while permitting controlled input-dependent calibration when shrinkage is relaxed.

Significance. If the orthogonality constraints provably restore identifiability and the nesting property holds, the IBFU construction would supply a principled way to regularize calibration parameters around best estimates while extending KOH to input-dependent settings. This could be useful in sparse-data regimes where standard KOH exhibits confounding pathologies. The explicit use of conservative hyperpriors to control the complexity of the correction GP is a potentially valuable modeling device.

major comments (2)
  1. [Abstract (model formulation)] Abstract (model formulation paragraph): the orthogonality constraints are asserted to mitigate confounding between the parameter correction GP and the additive discrepancy GP, yet no explicit mathematical form (inner-product conditions, projection operators, or modified likelihood) or derivation showing that the joint posterior factors or that identifiability is restored is supplied. This is load-bearing for the central claim that IBFU is both more universal and more tightly constrained than KOH.
  2. [Abstract (shrinkage and nesting claim)] Abstract (shrinkage and nesting claim): the statement that strong complexity shrinkage forces the mean parameter correction to remain flat (recovering KOH) while relaxing it permits input-dependent calibration in a controlled manner lacks either a formal proof of the nesting property or simulation studies demonstrating recovery of input-dependent parameters versus KOH in sparse-data settings. Without such verification the universality assertion cannot be assessed.
minor comments (1)
  1. [Abstract] The abstract is information-dense; separating the description of the model components from the claimed advantages would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We appreciate the referee's insightful comments highlighting key aspects of our IBFU framework. We provide point-by-point responses below, noting where revisions will be made to strengthen the presentation.

read point-by-point responses

  1. Referee: Abstract (model formulation paragraph): the orthogonality constraints are asserted to mitigate confounding between the parameter correction GP and the additive discrepancy GP, yet no explicit mathematical form (inner-product conditions, projection operators, or modified likelihood) or derivation showing that the joint posterior factors or that identifiability is restored is supplied. This is load-bearing for the central claim that IBFU is both more universal and more tightly constrained than KOH.

    Authors: While the abstract is necessarily concise, the full manuscript provides the mathematical details in Section 3: the orthogonality is enforced via the condition that the inner product of the parameter correction GP and the discrepancy GP is zero, implemented through a modified joint covariance that projects out the confounding components. The derivation that this restores identifiability is in Appendix A. We will revise the abstract to briefly reference these constraints and their purpose. revision: partial

  2. Referee: Abstract (shrinkage and nesting claim): the statement that strong complexity shrinkage forces the mean parameter correction to remain flat (recovering KOH) while relaxing it permits input-dependent calibration in a controlled manner lacks either a formal proof of the nesting property or simulation studies demonstrating recovery of input-dependent parameters versus KOH in sparse-data settings. Without such verification the universality assertion cannot be assessed.

    Authors: The nesting follows from the model construction: strong shrinkage priors on the correction GP variance and lengthscale force the correction to a constant, exactly recovering KOH as a special case. This is formalized in the hierarchical prior setup described in Section 2.2. The numerical studies in Section 4 show the behavior under varying shrinkage levels in sparse settings. We will update the abstract to cite the relevant section for the nesting property. revision: partial

Circularity Check

0 steps flagged

No circularity: IBFU model is a self-contained modeling proposal with external anchors

full rationale

The paper introduces the IBFU formalism as an explicit Bayesian modeling choice: calibration parameters are defined as best-estimate plus input-dependent GP correction under shrinkage hyperpriors, with an additive discrepancy GP and orthogonality constraints imposed by construction. This nests KOH only in the strong-shrinkage limit as a designed limiting case, not by reducing any prediction or uniqueness claim to a fitted input or self-citation. No equations or steps in the provided text equate a derived quantity to its own inputs by definition, and no load-bearing premise relies on prior work by the same authors. The construction is therefore independent of the target claims and receives the default non-circularity finding.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 1 invented entities

The proposal rests on standard GP modeling assumptions plus new modeling choices (parameter correction GP, orthogonality constraints, conservative hyperpriors) that are introduced without independent external evidence in the abstract.

free parameters (2)
  • shrinkage hyperpriors
    Conservative hyperpriors on GP complexity that force the mean parameter correction to remain flat unless data support variation.
  • best estimate values
    Fixed anchors around which the parameter corrections are defined.
axioms (2)
  • domain assumption Gaussian process priors are appropriate for both discrepancy and parameter corrections
    Invoked throughout the model description as the representation for functional components.
  • ad hoc to paper Orthogonality constraints successfully separate the parameter correction GP from the additive discrepancy GP
    Introduced to mitigate confounding but not derived from first principles in the abstract.
invented entities (1)
  • parameter correction GP no independent evidence
    purpose: To represent input-dependent adjustments to calibration parameters
    New component introduced to extend beyond fixed-parameter KOH.

pith-pipeline@v0.9.1-grok · 5814 in / 1349 out tokens · 25015 ms · 2026-06-29T05:40:41.609383+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

3 extracted references · 3 canonical work pages · 1 internal anchor

  1. [1]

    Journal of the Ameri- can Statistical Association111(514), 585–599 (2016).https://doi.org/10.1080/ 01621459.2016.11416855

    ISSN 0162-1459, 1537-274X. doi: 10.1080/01621459.2016.1211016. URLhttps://www.tandfonline.com/doi/full/10.1080/ 01621459.2016.1211016. Matthew Plumlee. Computer Model Calibration with Confidence and Consistency.Journal of the Royal Statistical Society Series B: Statistical Methodology, 81(3):519–545, July 2019. ISSN 1369-7412, 1467-9868. doi: 10.1111/rssb...

    work page doi:10.1080/01621459.2016.1211016 2016

  2. [2]

    Penalising model component complexity: A principled, practical approach to constructing priors

    ISSN 0030-1299, 1600-0706. doi: 10.1111/oik.05985. URLhttps://nsojournals.onlinelibrary.wiley.com/doi/10.1111/ oik.05985. Jenny Brynjarsdottir and Anthony O’Hagan. Learning about physical parameters: the importance of model discrepancy.Inverse Problems, 30 (11):114007,November2014. ISSN0266-5611,1361-6420. doi:10.1088/0266-5611/30/11/114007. URLhttps://io...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1111/oik.05985 2014

  3. [3]

    URLhttps://www.degruyterbrill.com/document/doi/10.1515/1544-6115.1803/html

    doi: 10.1515/1544-6115.1803. URLhttps://www.degruyterbrill.com/document/doi/10.1515/1544-6115.1803/html. Juho Piironen and Aki Vehtari. On the hyperprior choice for the global shrinkage parameter in the horseshoe prior. Liam Myhill. Caked-up. https://github.com/MUEXLY/caked-up, 2026. RobertTibshirani. Regressionshrinkageandselectionviathelasso. 58(1):267–...

    work page doi:10.1515/1544-6115.1803 2026