Journal of Computational Physics Submission Guide

Before formatting, write each central claim in the left column of a private audit table. For every claim, record the comparison that could weaken it, the plotted quantity that tests it, and the regime where it may fail.

In our editorial analysis, the recurring failure pattern is a benchmark substitution: a paper reports that its method wins one metric, then concludes that it is more useful overall. The correction is to keep accuracy, stability, runtime, memory, robustness, and interpretability separate until the evidence supports combining them.

Verification asks whether the implementation solves the equations as intended. Validation asks whether those equations and assumptions represent the physical phenomenon well enough for the stated use. A manuscript that compares only against its own simulated data may verify code behavior without validating the physical claim.

A useful evidence sequence is:

1. define the governing problem, discretization, constraints, and evaluation quantities;
2. verify against an analytic solution, manufactured solution, or trusted reference where possible;
3. establish convergence or error behavior under controlled refinement;
4. compare with credible baselines under matched conditions;
5. validate against experimental or observational evidence when the claim requires it;
6. expose the regime where performance or physical fidelity degrades.

The last item is information, not a weakness to hide. A failure map often gives a JCP reader more decision value than another average-performance number.

Suppose a paper introduces a neural surrogate that predicts a flow field much faster than a conventional solver. The first draft compares mean error on randomly split simulations. That does not yet establish the broad efficiency claim: nearby parameter samples can leak regime information, training cost may be excluded, conservation may drift, and the model may fail outside the training envelope.

The stronger submission separates inference speed from total computational cost, tests held-out physical regimes, reports conservation residuals, compares at matched accuracy, and shows where the surrogate should hand control back to a numerical solver. The resulting claim is narrower but more useful: the surrogate accelerates a defined class of evaluations within a documented operating envelope.

1. Confirm the current article type and scope on the official journal page.
2. Reconcile equations, symbols, units, boundary conditions, algorithms, figures, tables, and supplementary files.
3. Make the data, code, parameter, and software-environment statements match what can actually be accessed.
4. Check author, funding, conflict, ethics, and contributor declarations against the live guide.
5. Generate the submission PDF and inspect equations, fonts, vector graphics, links, and supplementary references.
6. Enter through the submission route linked by the official journal page rather than an old bookmark.

Do not infer a JCP requirement from another Elsevier journal. Shared publisher policies can support the package, but the live JCP guide controls journal-specific instructions.

Discretization result presented as physical discovery. A numerical artifact can look like a new regime when mesh, timestep, solver tolerance, or boundary treatment changes. Show refinement behavior and, where possible, an independent formulation or reference solution. If the phenomenon disappears under a reasonable numerical choice, the manuscript should explain that before calling it physics.

Efficiency claim with unmatched quality. A method is not faster in the decision-relevant sense when it reaches lower accuracy, excludes setup cost, or runs on different hardware without normalization. Record wall time, memory, training or preprocessing cost, convergence target, and the error level at which methods are compared. A Pareto view often tells the truth more clearly than one runtime ratio.

Reproducibility deferred to unavailable code. Code can support a paper, but it cannot replace algorithmic detail, parameter definitions, or the mathematical problem. State what another specialist needs to implement the method independently. If licensing or data restrictions prevent full release, describe the boundary and provide the most useful inspectable alternative.

Questions a reviewer should be able to answer

1. Which physical or computational obstacle does the method remove?
2. Which comparison most directly tests that claim?
3. What regime causes the method to fail or lose its advantage?
4. Can another team reconstruct the central result from the paper and declared artifacts?
5. Does the conclusion distinguish verified numerical behavior from validated physical inference?

These questions create a compact rebuttal stress test. If an answer depends on “the reviewer will understand,” add the missing definition, figure, table, or caveat to the manuscript. If it depends on a future experiment or unreleased artifact, narrow the present claim.