Chapter map

1. 1. Base mesh
2. 2. Feature
3. 3. Indicator
4. 4. Refine
5. 5. Count
6. 6. Uniform baseline
7. 7. Saving
8. 8. Prolongation
9. 9. Restriction
10. 10. Flux match
11. 11. Time
12. 12. Move feature
13. 13. Convergence
14. 14. Conclusion

Visit the How Mathematics Works hub for the wider series and its connections to education, reasoning, science, computing, design and careers.

---

1. Base mesh

The useful question is not only how to calculate, but what the calculation assumes. Divide [0,1] into four cells.

A concrete situation

Each width=.25.

What the mathematics does not license

Record coordinates. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Draw mesh. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

2. Feature

A small example makes the mechanism visible. Place a sharp transition near .5.

A concrete situation

Middle cells are candidates.

What the mathematics does not license

Define indicator. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record feature. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

3. Indicator

This is where modelling choices begin to affect evidence. Flag two middle cells.

A concrete situation

Decision is evidence-dependent.

What the mathematics does not license

State threshold. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record flags. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

4. Refine

The central mathematical idea is concrete. Bisect each flagged cell.

A concrete situation

Four fine cells of width .125.

What the mathematics does not license

Maintain nesting. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record mesh. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

5. Count

Here the geometry becomes operational. Add two coarse outer cells.

A concrete situation

Adaptive total=6.

What the mathematics does not license

Count leaves only. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record count. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

6. Uniform baseline

This step deserves a careful pause. Refine all base cells.

A concrete situation

Uniform total=8.

What the mathematics does not license

Same smallest width. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record baseline. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

7. Saving

The useful question is not only how to calculate, but what the calculation assumes. Compare 6 with 8.

A concrete situation

Two cells saved in 1D example.

What the mathematics does not license

Not an accuracy claim. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record ratio. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

8. Prolongation

A small example makes the mechanism visible. Fill fine data from coarse representation.

A concrete situation

Interpolation order matters.

What the mathematics does not license

Avoid new extrema. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record rule. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

9. Restriction

This is where modelling choices begin to affect evidence. Return fine averages to coarse level.

A concrete situation

Use conservative averaging.

What the mathematics does not license

Track weights. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record rule. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

10. Flux match

The central mathematical idea is concrete. Compare coarse and summed fine fluxes.

A concrete situation

Mismatch requires correction.

What the mathematics does not license

Use refluxing when eligible. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record balance. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

11. Time

Here the geometry becomes operational. Reduce fine-level dt.

A concrete situation

Subcycling may synchronize at coarse step.

What the mathematics does not license

Check CFL. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record schedule. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

12. Move feature

This step deserves a careful pause. Shift transition.

A concrete situation

Regrid or add buffer.

What the mathematics does not license

Avoid lag. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record test. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

13. Convergence

The useful question is not only how to calculate, but what the calculation assumes. Compare with uniform fine reference.

A concrete situation

Measure target error and cost.

What the mathematics does not license

Indicator needs evidence. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record table. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

14. Conclusion

A small example makes the mechanism visible. State mesh,ratio,indicator,transfer,synchronization and cost.

A concrete situation

Worked adaptive count=6 versus 8.

What the mathematics does not license

Efficiency is conditional. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Write conclusion. Then explain the result to someone who has not seen the formula. If the explanation needs a hidden assumption, add that assumption to the model record.

---

A seven-day practice lab

1. Day 1: define the decision, observation unit, X variables and Y or reference condition.
2. Day 2: reproduce the smallest worked example and check every dimension, unit and sign.
3. Day 3: build a transparent baseline and write what it assumes.
4. Day 4: fit the main method inside a training-only pipeline; record every parameter.
5. Day 5: challenge one assumption with a deliberately difficult dataset or operating regime.
6. Day 6: compare validation metrics with the practical workload created by errors.
7. Day 7: present one supported conclusion, one unresolved uncertainty and one safe next experiment.

Questions for a student, parent or mentor

• Which part of the model is a mathematical definition, and which part is a choice?
• What small calculation would catch a sign, scale or dimension error?
• What evidence comes from training, validation and the untouched test?
• What benign condition could look like a fault or prediction failure?
• How would the answer change for another response, asset, school project or time period?
• What should a human review before acting on the score?

A method-specific audit before deployment

Adaptive mesh refinement concentrates small cells where a numerical solution needs detail and keeps larger cells where the solution is smooth. The mathematics decides where resolution is valuable and how neighbouring levels exchange information.

Begin with four equal cells on [0,1], each of width 0.25. Suppose an error indicator flags the two middle cells around a sharp transition.

Bisect only those two cells. The new widths are 0.25, 0.125, 0.125, 0.125, 0.125 and 0.25: six cells provide twice the local spatial resolution near the transition.

A uniform refinement to the same smallest width would use eight cells. In this simple one-dimensional count, local refinement saves two cells; in two or three dimensions the potential savings can be much larger.

Cell count alone is not an accuracy proof. The indicator must correlate with the error or feature of interest, and the coarse-fine interface must transmit fluxes, boundary data and time updates consistently.

Berger and Oliger’s 1984 algorithm used nested component grids, Richardson-type truncation-error estimates and refinement in both space and time for hyperbolic equations.

Did You Know? Fine spatial cells often require smaller explicit time steps. Local time stepping or subcycling can save work, but synchronization and conservation at level boundaries become part of the algorithm.

An error estimator approximates a mathematical error; a feature sensor may instead target gradients, shocks or vorticity. They answer different questions and should not be presented as interchangeable evidence.

Refinement thresholds create a decision rule. Too low a threshold expands fine regions and cost; too high a threshold misses structure. Hysteresis can reduce repeated refine-coarsen oscillation.

Proper nesting keeps fine grids inside valid coarse regions and supplies ghost-cell data. Interpolation from coarse to fine and restriction from fine to coarse need stated orders and conservation properties.

Refluxing corrects coarse and fine flux mismatches in conservative finite-volume AMR. Without it, a visually smooth solution can quietly gain or lose mass across a level boundary.

Students can mark cells by a gradient indicator, refine them, recount degrees of freedom and then compare with a uniformly refined reference. The exercise separates efficiency from accuracy.

Applications include weather, astrophysics, combustion, shock waves, groundwater and structural singularities. The refinement rule must follow the quantity that matters to the application.

A moving feature can outrun a static refined patch. Regridding frequency, buffer zones and prediction determine whether resolution stays ahead of the physics.

For parents and students, the transferable idea is selective attention with accountability: spend effort where evidence says it matters, and keep checks at the borders between detailed and coarse work.

A responsible report states the base mesh, refinement ratio, indicator, thresholds, nesting rules, interpolation, restriction, synchronization, time stepping, conservation checks, convergence and cost.

---

Frequently asked questions

What is adaptive mesh refinement?

A method that adds resolution selectively where an estimator or feature indicator identifies need.

Why not refine everywhere?

Uniform refinement can spend most degrees of freedom in smooth regions.

What is an error indicator?

A computable signal used to decide which cells or elements may need refinement.

Why are coarse-fine interfaces difficult?

Interpolation, fluxes and time levels must remain consistent across different resolutions.

Does fewer cells prove equal accuracy?

No. Accuracy must be measured against reference solutions or independent evidence.

What must be reported?

Base mesh, ratios, indicator, thresholds, transfer rules, synchronization, conservation, convergence and cost.

---

Primary source and fact-check note

Marsha J. Berger and Joseph Oliger, “Adaptive Mesh Refinement for Hyperbolic Partial Differential Equations,” Journal of Computational Physics 53, 1984. Definitions and core claims in this article were checked against this primary publication. Calculations shown here were independently recomputed. Examples are educational illustrations; they are not reported industrial performance results.

Useful next reading

• Finite volume method
• Finite difference method
• Computational mathematics hub
• How Mathematics Works
• Study and learning methods

Keep the options open. Mathematics education supports clearer questions, transferable problem-solving skills and informed choices across science, engineering, computing, AI, statistics, design and many other careers. It does not guarantee intelligence, income, admission or employment; those outcomes depend on far more than one subject or model.

Method audit: equations, structure and evidence

Indicator audit

State whether the trigger estimates error or merely detects a feature.

Interface audit

Check interpolation, restriction, flux matching and time synchronization.

Efficiency audit

Compare error versus cost with a uniform reference.

Motion audit

Verify that moving features remain inside adequately refined regions.

Mechanism workshop: calculate, challenge and transfer

Mesh drawing

Label every coarse and fine cell edge.

Indicator table

Calculate a gradient or Richardson signal per cell.

Threshold sweep

Count cells and errors at several thresholds.

Transfer test

Prolong and restrict a constant and linear field.

Flux test

Measure coarse-fine conservation before and after correction.

Time schedule

Draw fine substeps inside one coarse step.

Moving pulse

Regrid as a peak crosses a level boundary.

Cost curve

Plot target error against cell-updates and time.

Student explanation

Explain why six cells can target detail better than four.

Parent explanation

Describe selective effort with border checks.

Evidence notebook: reconstruct every step

Evidence notebook 1: Base mesh

Step 1 isolates base mesh. Divide [0,1] into four cells. The checked result is: Each width=.25. Keep the equation, units, node or element order, face orientation, basis convention, time level and precision beside every value. The safeguard is: Record coordinates. Draw mesh. Recalculate after changing exactly one parameter and explain whether the change arises from spatial approximation, time integration, geometry preservation, refinement logic, numerical flux, rounding or a different physical model. Ask a second reader to reproduce the arithmetic and finish with the narrowest claim the evidence supports.

Evidence notebook 2: Feature

Step 2 isolates feature. Place a sharp transition near .5. The checked result is: Middle cells are candidates. Keep the equation, units, node or element order, face orientation, basis convention, time level and precision beside every value. The safeguard is: Define indicator. Record feature. Recalculate after changing exactly one parameter and explain whether the change arises from spatial approximation, time integration, geometry preservation, refinement logic, numerical flux, rounding or a different physical model. Ask a second reader to reproduce the arithmetic and finish with the narrowest claim the evidence supports.

Evidence notebook 3: Indicator

Step 3 isolates indicator. Flag two middle cells. The checked result is: Decision is evidence-dependent. Keep the equation, units, node or element order, face orientation, basis convention, time level and precision beside every value. The safeguard is: State threshold. Record flags. Recalculate after changing exactly one parameter and explain whether the change arises from spatial approximation, time integration, geometry preservation, refinement logic, numerical flux, rounding or a different physical model. Ask a second reader to reproduce the arithmetic and finish with the narrowest claim the evidence supports.

Evidence notebook 4: Refine

Step 4 isolates refine. Bisect each flagged cell. The checked result is: Four fine cells of width .125. Keep the equation, units, node or element order, face orientation, basis convention, time level and precision beside every value. The safeguard is: Maintain nesting. Record mesh. Recalculate after changing exactly one parameter and explain whether the change arises from spatial approximation, time integration, geometry preservation, refinement logic, numerical flux, rounding or a different physical model. Ask a second reader to reproduce the arithmetic and finish with the narrowest claim the evidence supports.

Evidence notebook 5: Count

Step 5 isolates count. Add two coarse outer cells. The checked result is: Adaptive total=6. Keep the equation, units, node or element order, face orientation, basis convention, time level and precision beside every value. The safeguard is: Count leaves only. Record count. Recalculate after changing exactly one parameter and explain whether the change arises from spatial approximation, time integration, geometry preservation, refinement logic, numerical flux, rounding or a different physical model. Ask a second reader to reproduce the arithmetic and finish with the narrowest claim the evidence supports.

Evidence notebook 6: Uniform baseline

Step 6 isolates uniform baseline. Refine all base cells. The checked result is: Uniform total=8. Keep the equation, units, node or element order, face orientation, basis convention, time level and precision beside every value. The safeguard is: Same smallest width. Record baseline. Recalculate after changing exactly one parameter and explain whether the change arises from spatial approximation, time integration, geometry preservation, refinement logic, numerical flux, rounding or a different physical model. Ask a second reader to reproduce the arithmetic and finish with the narrowest claim the evidence supports.

Evidence notebook 7: Saving

Step 7 isolates saving. Compare 6 with 8. The checked result is: Two cells saved in 1D example. Keep the equation, units, node or element order, face orientation, basis convention, time level and precision beside every value. The safeguard is: Not an accuracy claim. Record ratio. Recalculate after changing exactly one parameter and explain whether the change arises from spatial approximation, time integration, geometry preservation, refinement logic, numerical flux, rounding or a different physical model. Ask a second reader to reproduce the arithmetic and finish with the narrowest claim the evidence supports.

Evidence notebook 8: Prolongation

Step 8 isolates prolongation. Fill fine data from coarse representation. The checked result is: Interpolation order matters. Keep the equation, units, node or element order, face orientation, basis convention, time level and precision beside every value. The safeguard is: Avoid new extrema. Record rule. Recalculate after changing exactly one parameter and explain whether the change arises from spatial approximation, time integration, geometry preservation, refinement logic, numerical flux, rounding or a different physical model. Ask a second reader to reproduce the arithmetic and finish with the narrowest claim the evidence supports.

Evidence notebook 9: Restriction

Step 9 isolates restriction. Return fine averages to coarse level. The checked result is: Use conservative averaging. Keep the equation, units, node or element order, face orientation, basis convention, time level and precision beside every value. The safeguard is: Track weights. Record rule. Recalculate after changing exactly one parameter and explain whether the change arises from spatial approximation, time integration, geometry preservation, refinement logic, numerical flux, rounding or a different physical model. Ask a second reader to reproduce the arithmetic and finish with the narrowest claim the evidence supports.

Evidence notebook 10: Flux match

Step 10 isolates flux match. Compare coarse and summed fine fluxes. The checked result is: Mismatch requires correction. Keep the equation, units, node or element order, face orientation, basis convention, time level and precision beside every value. The safeguard is: Use refluxing when eligible. Record balance. Recalculate after changing exactly one parameter and explain whether the change arises from spatial approximation, time integration, geometry preservation, refinement logic, numerical flux, rounding or a different physical model. Ask a second reader to reproduce the arithmetic and finish with the narrowest claim the evidence supports.

Evidence notebook 11: Time

Step 11 isolates time. Reduce fine-level dt. The checked result is: Subcycling may synchronize at coarse step. Keep the equation, units, node or element order, face orientation, basis convention, time level and precision beside every value. The safeguard is: Check CFL. Record schedule. Recalculate after changing exactly one parameter and explain whether the change arises from spatial approximation, time integration, geometry preservation, refinement logic, numerical flux, rounding or a different physical model. Ask a second reader to reproduce the arithmetic and finish with the narrowest claim the evidence supports.

Evidence notebook 12: Move feature

Step 12 isolates move feature. Shift transition. The checked result is: Regrid or add buffer. Keep the equation, units, node or element order, face orientation, basis convention, time level and precision beside every value. The safeguard is: Avoid lag. Record test. Recalculate after changing exactly one parameter and explain whether the change arises from spatial approximation, time integration, geometry preservation, refinement logic, numerical flux, rounding or a different physical model. Ask a second reader to reproduce the arithmetic and finish with the narrowest claim the evidence supports.

Evidence notebook 13: Convergence

Step 13 isolates convergence. Compare with uniform fine reference. The checked result is: Measure target error and cost. Keep the equation, units, node or element order, face orientation, basis convention, time level and precision beside every value. The safeguard is: Indicator needs evidence. Record table. Recalculate after changing exactly one parameter and explain whether the change arises from spatial approximation, time integration, geometry preservation, refinement logic, numerical flux, rounding or a different physical model. Ask a second reader to reproduce the arithmetic and finish with the narrowest claim the evidence supports.

Evidence notebook 14: Conclusion

Step 14 isolates conclusion. State mesh,ratio,indicator,transfer,synchronization and cost. The checked result is: Worked adaptive count=6 versus 8. Keep the equation, units, node or element order, face orientation, basis convention, time level and precision beside every value. The safeguard is: Efficiency is conditional. Write conclusion. Recalculate after changing exactly one parameter and explain whether the change arises from spatial approximation, time integration, geometry preservation, refinement logic, numerical flux, rounding or a different physical model. Ask a second reader to reproduce the arithmetic and finish with the narrowest claim the evidence supports.

Failure cases and responsible recovery

Gradient-is-error

A feature sensor is not an estimator. Build a small counterexample, restore the missing condition and recalculate before making the claim.

Cell-count proof

Efficiency needs equal-error comparison. Build a small counterexample, restore the missing condition and recalculate before making the claim.

Interface silence

Coarse-fine mismatch can break conservation. Build a small counterexample, restore the missing condition and recalculate before making the claim.

No time refinement

Explicit fine cells may violate CFL. Build a small counterexample, restore the missing condition and recalculate before making the claim.

Chattering mesh

Threshold design needs hysteresis or buffers. Build a small counterexample, restore the missing condition and recalculate before making the claim.

Static patch

Moving features can escape refinement. Build a small counterexample, restore the missing condition and recalculate before making the claim.

No uniform baseline

Savings cannot be assessed. Build a small counterexample, restore the missing condition and recalculate before making the claim.

Pretty detail

Local resolution does not validate physics. Build a small counterexample, restore the missing condition and recalculate before making the claim.

Verified case study: fourteen steps from equation to decision

1. Define the continuous model

Write the equation, variables, units, domain and initial or boundary data.

2. Identify preserved structure

Name conservation, dissipation, symplectic geometry, smoothness or localisation that should guide the discretisation.

3. Choose representation

Justify nodes, elements, local polynomials, nested grids or phase variables.

4. Derive one local rule

Connect a matrix row, canonical substep, refinement decision or face flux to the governing mathematics.

5. Calculate by hand

Use the smallest nontrivial example and keep exact values where practical.

6. Check signs and units

Audit orientation, dimensions, time levels and indexing.

7. Test an invariant

Measure mass, phase area, energy behaviour, boundary values or another eligible diagnostic.

8. Trigger a known failure

Use a bad time step, escaped feature, wrong trace or broken interface transfer.

9. Refine systematically

Vary one of space, time, polynomial degree or threshold while controlling the others.

10. Compare a baseline

Use an exact, manufactured, uniform-grid or independently implemented reference.

11. Count cost

Record degrees of freedom, cell-updates, matrix solves, memory and elapsed time.

12. Validate the model

Compare the application output with independent observations or accepted benchmarks.

13. Communicate limits

State what the result supports and what remains untested.

14. Preserve evidence

Save equations, data, code, meshes, parameters, logs and corrected calculations.

Learner lab: ten checks for transferable understanding

Lab question 1

Create a reproducible mini-investigation of indicator, threshold, nested mesh, transfer, refluxing, subcycling and cost. For task 1, state one prediction, calculate or simulate a transparent case, record units and conventions, compare an independent baseline, identify one failure mode, and explain the result once for a student and once for a parent. End with one practical next step and one claim the evidence does not support.

Lab question 2

Create a reproducible mini-investigation of indicator, threshold, nested mesh, transfer, refluxing, subcycling and cost. For task 2, state one prediction, calculate or simulate a transparent case, record units and conventions, compare an independent baseline, identify one failure mode, and explain the result once for a student and once for a parent. End with one practical next step and one claim the evidence does not support.

Lab question 3

Create a reproducible mini-investigation of indicator, threshold, nested mesh, transfer, refluxing, subcycling and cost. For task 3, state one prediction, calculate or simulate a transparent case, record units and conventions, compare an independent baseline, identify one failure mode, and explain the result once for a student and once for a parent. End with one practical next step and one claim the evidence does not support.

Lab question 4

Create a reproducible mini-investigation of indicator, threshold, nested mesh, transfer, refluxing, subcycling and cost. For task 4, state one prediction, calculate or simulate a transparent case, record units and conventions, compare an independent baseline, identify one failure mode, and explain the result once for a student and once for a parent. End with one practical next step and one claim the evidence does not support.

Lab question 5

Create a reproducible mini-investigation of indicator, threshold, nested mesh, transfer, refluxing, subcycling and cost. For task 5, state one prediction, calculate or simulate a transparent case, record units and conventions, compare an independent baseline, identify one failure mode, and explain the result once for a student and once for a parent. End with one practical next step and one claim the evidence does not support.

Lab question 6

Create a reproducible mini-investigation of indicator, threshold, nested mesh, transfer, refluxing, subcycling and cost. For task 6, state one prediction, calculate or simulate a transparent case, record units and conventions, compare an independent baseline, identify one failure mode, and explain the result once for a student and once for a parent. End with one practical next step and one claim the evidence does not support.

Lab question 7

Create a reproducible mini-investigation of indicator, threshold, nested mesh, transfer, refluxing, subcycling and cost. For task 7, state one prediction, calculate or simulate a transparent case, record units and conventions, compare an independent baseline, identify one failure mode, and explain the result once for a student and once for a parent. End with one practical next step and one claim the evidence does not support.

Lab question 8

Create a reproducible mini-investigation of indicator, threshold, nested mesh, transfer, refluxing, subcycling and cost. For task 8, state one prediction, calculate or simulate a transparent case, record units and conventions, compare an independent baseline, identify one failure mode, and explain the result once for a student and once for a parent. End with one practical next step and one claim the evidence does not support.

Lab question 9

Create a reproducible mini-investigation of indicator, threshold, nested mesh, transfer, refluxing, subcycling and cost. For task 9, state one prediction, calculate or simulate a transparent case, record units and conventions, compare an independent baseline, identify one failure mode, and explain the result once for a student and once for a parent. End with one practical next step and one claim the evidence does not support.

Lab question 10

Create a reproducible mini-investigation of indicator, threshold, nested mesh, transfer, refluxing, subcycling and cost. For task 10, state one prediction, calculate or simulate a transparent case, record units and conventions, compare an independent baseline, identify one failure mode, and explain the result once for a student and once for a parent. End with one practical next step and one claim the evidence does not support.