Chapter map

1. 1. Nodes
2. 2. Evaluation
3. 3. Windows
4. 4. Basis
5. 5. Moment matrix
6. 6. Determinant
7. 7. Inverse
8. 8. Shape zero
9. 9. Shape one
10. 10. Constant test
11. 11. Linear test
12. 12. Derivative
13. 13. Perturbation
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. Nodes

This is where modelling choices begin to affect evidence. Set x0=0 and x1=1.

A concrete situation

Two-node one-dimensional cloud.

What the mathematics does not license

State support. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record coordinates. 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. Evaluation

The central mathematical idea is concrete. Set x=.25.

A concrete situation

Point lies between nodes.

What the mathematics does not license

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

Try it, check it, transfer it

Record x. 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. Windows

Here the geometry becomes operational. Use w0=1,w1=.5.

A concrete situation

Both nodes contribute.

What the mathematics does not license

Illustrative raw weights. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record values. 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. Basis

This step deserves a careful pause. Use p=[1,x].

A concrete situation

Target constant and linear reproduction.

What the mathematics does not license

Column convention matters. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record basis. 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. Moment matrix

The useful question is not only how to calculate, but what the calculation assumes. Sum w_i p_i p_i transpose.

A concrete situation

M=[[1.5,.5],[.5,.5]].

What the mathematics does not license

Include every neighbour once. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record matrix. 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. Determinant

A small example makes the mechanism visible. Compute .75-.25.

A concrete situation

det M=.5.

What the mathematics does not license

Nonzero in this cloud. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record conditioning too. 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. Inverse

This is where modelling choices begin to affect evidence. Invert the 2×2 matrix.

A concrete situation

M inverse=[[1,-1],[-1,3]].

What the mathematics does not license

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

Try it, check it, transfer it

Record inverse. 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. Shape zero

The central mathematical idea is concrete. Apply correction at x0.

A concrete situation

phi0=.75.

What the mathematics does not license

Use evaluation basis. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record value. 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. Shape one

Here the geometry becomes operational. Apply correction at x1.

A concrete situation

phi1=.25.

What the mathematics does not license

Include w1=.5. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record value. 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. Constant test

This step deserves a careful pause. Add corrected shapes.

A concrete situation

Sum=1.

What the mathematics does not license

Partition of unity. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record check. 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. Linear test

The useful question is not only how to calculate, but what the calculation assumes. Sum phi_i x_i.

A concrete situation

Coordinate=.25.

What the mathematics does not license

First-order consistency. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record check. 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. Derivative

A small example makes the mechanism visible. Differentiate kernel and correction.

A concrete situation

Linear derivative should reproduce one.

What the mathematics does not license

Full derivative required. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record result. 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. Perturbation

This is where modelling choices begin to affect evidence. Move x1 or reduce support.

A concrete situation

Moment conditioning changes.

What the mathematics does not license

Recompute all terms. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record sensitivity. 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

The central mathematical idea is concrete. State support, correction, integration and boundaries.

A concrete situation

Worked shapes .75,.25 reproduce x=.25.

What the mathematics does not license

Two-node teaching cloud. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Write narrow claim. 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

The reproducing kernel particle method builds mesh-free approximation functions from a window or kernel and a correction that enforces polynomial consistency. The correction is the mathematical bridge between a convenient local weighting rule and the ability to reproduce constants, linear fields or higher-order polynomials exactly.

Use a one-dimensional teaching cloud with nodes x0=0 and x1=1. At the evaluation point x=0.25, choose raw window weights w0=1 and w1=0.5 and the linear polynomial basis p(x)=[1,x]^T.

The moment matrix is the sum of w_i p(x_i)p(x_i)^T. Here it is [[1.5,0.5],[0.5,0.5]]. Its determinant is 1.5×0.5-0.5×0.5=0.5, so the matrix is invertible.

The inverse moment matrix is [[1,-1],[-1,3]]. Multiplying the evaluation basis [1,0.25] by this inverse and by the nodal basis and window gives corrected shape value φ0=0.75 at x0.

For x1, the same calculation gives φ1=0.25. The corrected values sum to one, so they reproduce a constant field exactly in this small cloud.

They also reproduce the coordinate: φ0x0+φ1×1=0.75×0+0.25×1=0.25, which equals the evaluation coordinate. This is first-order consistency made visible by a moment calculation.

Did You Know? Kernel normalisation alone may reproduce constants but fail to reproduce a linear field on an irregular cloud. RKPM’s correction uses local moments to restore the chosen polynomial completeness when the moment matrix is well conditioned.

The support radius determines which particles contribute to the moment matrix. Too few or poorly arranged neighbours can make that matrix singular or ill conditioned; excessively large support increases cost and may smooth away local variation.

Derivatives of corrected shape functions include derivatives of the kernel and the correction. Differentiating only the raw window silently changes the approximation and can destroy the intended consistency.

A Galerkin formulation still needs numerical integration. Background cells, nodal integration or stabilised conforming integration are separate choices; calling a method mesh-free does not mean every stage is free of integration structure.

Essential boundary conditions are less direct when shape functions do not interpolate nodal values. Transformation methods, penalty terms, Lagrange multipliers or coupling techniques must be stated and checked.

Large deformation changes neighbour geometry, moment matrices and support. Updating the correction consistently matters for objectivity, stability and computational cost.

Students can form a small moment matrix, invert it, compute corrected shape values and test constant and linear reproduction. Perturbing a node shows how conditioning and support alter the calculation.

Applications include structural dynamics, impact, forming and problems with large deformation or evolving geometry. RKPM does not remove the need for constitutive modelling, stable time integration and validation.

For parents and students, the transferable lesson is that a smart correction can make a local weighted average obey exact algebraic promises before it is trusted on a difficult problem.

A responsible RKPM report states kernel and support, polynomial basis, moment matrix, correction, derivatives, integration, boundary enforcement, stabilisation, neighbour update, time integration, completeness tests, convergence and validation.

---

Frequently asked questions

What is RKPM?

A mesh-free approximation method that corrects kernel functions to reproduce chosen polynomial fields.

What is the moment matrix?

A local weighted matrix of polynomial basis products whose inverse defines the correction.

What does consistency mean?

The approximation exactly reproduces specified polynomial fields, such as constants and linear coordinates.

Why can support fail?

Too few or badly arranged neighbours can make the moment matrix singular or ill conditioned.

Does mesh-free mean no numerical integration?

No. Many RKPM formulations still use background or stabilised integration.

What must be reported?

Kernel, support, basis, correction, derivatives, integration, boundaries, stabilisation, convergence and validation.

---

Primary source and fact-check note

W. K. Liu, S. Jun, S. Li, J. Adee and T. Belytschko, “Reproducing Kernel Particle Methods for Structural Dynamics,” International Journal for Numerical Methods in Engineering 38, 1995. 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

• Smoothed particle hydrodynamics
• Finite element method
• Peridynamics
• Numerical analysis hub
• How Mathematics Works

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: what must be measured

Moment ledger

Record neighbours, windows, basis, moment matrix and conditioning.

Reproduction ledger

Test constants, coordinates and requested polynomial order.

Derivative ledger

Differentiate the complete corrected approximation.

Weak-form ledger

State integration, boundaries, stabilisation and convergence.

Worked evidence notebook: reconstruct every number

Evidence notebook 1: Nodes

Set x0=0 and x1=1. Two-node one-dimensional cloud. Keep every unit, direction, coordinate, matrix entry, sign convention and rounding decision visible. State support. Record coordinates. Then change one input, recompute the affected mechanism and explain whether the difference comes from geometry, constitutive choice, kernel support, conditioning, boundary data, time integration or numerical precision. Preserve the calculation so another reader can reproduce the narrow conclusion.

Evidence notebook 2: Evaluation

Set x=.25. Point lies between nodes. Keep every unit, direction, coordinate, matrix entry, sign convention and rounding decision visible. State polynomial order. Record x. Then change one input, recompute the affected mechanism and explain whether the difference comes from geometry, constitutive choice, kernel support, conditioning, boundary data, time integration or numerical precision. Preserve the calculation so another reader can reproduce the narrow conclusion.

Evidence notebook 3: Windows

Use w0=1,w1=.5. Both nodes contribute. Keep every unit, direction, coordinate, matrix entry, sign convention and rounding decision visible. Illustrative raw weights. Record values. Then change one input, recompute the affected mechanism and explain whether the difference comes from geometry, constitutive choice, kernel support, conditioning, boundary data, time integration or numerical precision. Preserve the calculation so another reader can reproduce the narrow conclusion.

Evidence notebook 4: Basis

Use p=[1,x]. Target constant and linear reproduction. Keep every unit, direction, coordinate, matrix entry, sign convention and rounding decision visible. Column convention matters. Record basis. Then change one input, recompute the affected mechanism and explain whether the difference comes from geometry, constitutive choice, kernel support, conditioning, boundary data, time integration or numerical precision. Preserve the calculation so another reader can reproduce the narrow conclusion.

Evidence notebook 5: Moment matrix

Sum w_i p_i p_i transpose. M=[[1.5,.5],[.5,.5]]. Keep every unit, direction, coordinate, matrix entry, sign convention and rounding decision visible. Include every neighbour once. Record matrix. Then change one input, recompute the affected mechanism and explain whether the difference comes from geometry, constitutive choice, kernel support, conditioning, boundary data, time integration or numerical precision. Preserve the calculation so another reader can reproduce the narrow conclusion.

Evidence notebook 6: Determinant

Compute .75-.25. det M=.5. Keep every unit, direction, coordinate, matrix entry, sign convention and rounding decision visible. Nonzero in this cloud. Record conditioning too. Then change one input, recompute the affected mechanism and explain whether the difference comes from geometry, constitutive choice, kernel support, conditioning, boundary data, time integration or numerical precision. Preserve the calculation so another reader can reproduce the narrow conclusion.

Evidence notebook 7: Inverse

Invert the 2×2 matrix. M inverse=[[1,-1],[-1,3]]. Keep every unit, direction, coordinate, matrix entry, sign convention and rounding decision visible. Check multiplication. Record inverse. Then change one input, recompute the affected mechanism and explain whether the difference comes from geometry, constitutive choice, kernel support, conditioning, boundary data, time integration or numerical precision. Preserve the calculation so another reader can reproduce the narrow conclusion.

Evidence notebook 8: Shape zero

Apply correction at x0. phi0=.75. Keep every unit, direction, coordinate, matrix entry, sign convention and rounding decision visible. Use evaluation basis. Record value. Then change one input, recompute the affected mechanism and explain whether the difference comes from geometry, constitutive choice, kernel support, conditioning, boundary data, time integration or numerical precision. Preserve the calculation so another reader can reproduce the narrow conclusion.

Evidence notebook 9: Shape one

Apply correction at x1. phi1=.25. Keep every unit, direction, coordinate, matrix entry, sign convention and rounding decision visible. Include w1=.5. Record value. Then change one input, recompute the affected mechanism and explain whether the difference comes from geometry, constitutive choice, kernel support, conditioning, boundary data, time integration or numerical precision. Preserve the calculation so another reader can reproduce the narrow conclusion.

Evidence notebook 10: Constant test

Add corrected shapes. Sum=1. Keep every unit, direction, coordinate, matrix entry, sign convention and rounding decision visible. Partition of unity. Record check. Then change one input, recompute the affected mechanism and explain whether the difference comes from geometry, constitutive choice, kernel support, conditioning, boundary data, time integration or numerical precision. Preserve the calculation so another reader can reproduce the narrow conclusion.

Evidence notebook 11: Linear test

Sum phi_i x_i. Coordinate=.25. Keep every unit, direction, coordinate, matrix entry, sign convention and rounding decision visible. First-order consistency. Record check. Then change one input, recompute the affected mechanism and explain whether the difference comes from geometry, constitutive choice, kernel support, conditioning, boundary data, time integration or numerical precision. Preserve the calculation so another reader can reproduce the narrow conclusion.

Evidence notebook 12: Derivative

Differentiate kernel and correction. Linear derivative should reproduce one. Keep every unit, direction, coordinate, matrix entry, sign convention and rounding decision visible. Full derivative required. Record result. Then change one input, recompute the affected mechanism and explain whether the difference comes from geometry, constitutive choice, kernel support, conditioning, boundary data, time integration or numerical precision. Preserve the calculation so another reader can reproduce the narrow conclusion.

Evidence notebook 13: Perturbation

Move x1 or reduce support. Moment conditioning changes. Keep every unit, direction, coordinate, matrix entry, sign convention and rounding decision visible. Recompute all terms. Record sensitivity. Then change one input, recompute the affected mechanism and explain whether the difference comes from geometry, constitutive choice, kernel support, conditioning, boundary data, time integration or numerical precision. Preserve the calculation so another reader can reproduce the narrow conclusion.

Evidence notebook 14: Conclusion

State support, correction, integration and boundaries. Worked shapes .75,.25 reproduce x=.25. Keep every unit, direction, coordinate, matrix entry, sign convention and rounding decision visible. Two-node teaching cloud. Write narrow claim. Then change one input, recompute the affected mechanism and explain whether the difference comes from geometry, constitutive choice, kernel support, conditioning, boundary data, time integration or numerical precision. Preserve the calculation so another reader can reproduce the narrow conclusion.

Misconceptions, failure modes and recovery

Normalised kernels reproduce linears

Partition of unity alone guarantees only constants. Build the smallest counterexample, restore the missing condition and repeat the calculation before making a broader claim.

Moment inversion is always safe

Poor neighbour geometry can make it singular. Build the smallest counterexample, restore the missing condition and repeat the calculation before making a broader claim.

Raw kernel derivative is enough

The correction also depends on position. Build the smallest counterexample, restore the missing condition and repeat the calculation before making a broader claim.

Mesh-free means grid-free everywhere

Integration may use background cells. Build the smallest counterexample, restore the missing condition and repeat the calculation before making a broader claim.

Nodal values are interpolated

RKPM shape functions are generally non-interpolatory. Build the smallest counterexample, restore the missing condition and repeat the calculation before making a broader claim.

Large support is always safer

Cost and excessive smoothing can increase. Build the smallest counterexample, restore the missing condition and repeat the calculation before making a broader claim.

Consistency guarantees stability

Integration and time stepping still matter. Build the smallest counterexample, restore the missing condition and repeat the calculation before making a broader claim.

Smooth fields prove physics

Convergence and validation remain required. Build the smallest counterexample, restore the missing condition and repeat the calculation before making a broader claim.

A responsible eight-step decision path

1. Choose completeness

Name the polynomial fields to reproduce.

2. Choose support

Select kernel, dilation and neighbours.

3. Build moments

Form and inspect the local moment matrix.

4. Correct the kernel

Compute shape functions and full derivatives.

5. Discretise physics

Define weak form, integration and constitutive model.

6. Enforce boundaries

Use an eligible non-interpolatory treatment.

7. Refine

Vary points, support and integration consistently.

8. Validate

Use patch tests, convergence and physical evidence.

Learner lab: build transferable understanding

Moment matrix

Build M for a three-node irregular cloud. State a prediction, show the arithmetic, record a diagnostic, compare a baseline and finish with one justified conclusion plus one limit.

Constant test

Verify shape functions sum to one. State a prediction, show the arithmetic, record a diagnostic, compare a baseline and finish with one justified conclusion plus one limit.

Linear test

Verify coordinate reproduction. State a prediction, show the arithmetic, record a diagnostic, compare a baseline and finish with one justified conclusion plus one limit.

Conditioning

Move one neighbour until M becomes unstable. State a prediction, show the arithmetic, record a diagnostic, compare a baseline and finish with one justified conclusion plus one limit.

Support

Change the dilation radius and recount neighbours. State a prediction, show the arithmetic, record a diagnostic, compare a baseline and finish with one justified conclusion plus one limit.

Derivative

Reproduce the derivative of a linear field. State a prediction, show the arithmetic, record a diagnostic, compare a baseline and finish with one justified conclusion plus one limit.

Boundary

Compare penalty and transformation enforcement. State a prediction, show the arithmetic, record a diagnostic, compare a baseline and finish with one justified conclusion plus one limit.

Integration

Run a patch test with two quadrature choices. State a prediction, show the arithmetic, record a diagnostic, compare a baseline and finish with one justified conclusion plus one limit.

Student explanation

Explain correction of a weighted average. State a prediction, show the arithmetic, record a diagnostic, compare a baseline and finish with one justified conclusion plus one limit.

Parent explanation

Relate local evidence to an exact algebraic promise. State a prediction, show the arithmetic, record a diagnostic, compare a baseline and finish with one justified conclusion plus one limit.

Deeper transfer: from classroom mechanism to responsible use

Why the moment matrix matters

The moment matrix asks whether the local neighbour cloud contains enough independent geometric information to reconstruct the chosen polynomial basis. A nonzero determinant is useful in a two-dimensional teaching case, but practical codes should monitor condition numbers or stable factorizations rather than determinants alone.

Higher-order completeness

To reproduce quadratics, extend the polynomial basis with squared and mixed-coordinate terms. That increases the number and geometric diversity of neighbours required. Higher order can improve smooth accuracy but also enlarges support, raises cost and makes boundary truncation more demanding.

Partition of unity

When corrected shape functions sum to one, a constant field is reproduced. This property also supports consistent interpolation of rigid translations. It does not by itself guarantee exact linear strain, stable integration or conservation under a chosen time integrator.

Support near boundaries

A kernel centred near a boundary loses neighbours on one side. The correction can restore polynomial reproduction when the remaining moment matrix is adequate, but conditioning may deteriorate. Ghost particles, asymmetric support or special integration can change the response.

Nodal integration and stability

Direct nodal integration is attractive because it preserves the mesh-free spirit, yet under-integration can introduce zero-energy modes. Stabilised conforming nodal integration and related techniques add assumptions that need patch tests and energy checks.

Constitutive history

For large-deformation solids, particles may carry stress, plastic strain, damage and temperature. RKPM supplies approximation and gradients; it does not choose the material law. Objective stress updates and consistent tangents remain separate responsibilities.

Cost and neighbour updates

Moment matrices are local but must be assembled or updated for many evaluation points. Spatial search, matrix factorisation and derivative evaluation dominate cost. Reusing stale neighbours can save time while violating the current geometry after large motion.

Career connection

Computational mechanics researchers use RKPM in impact, forming and failure studies. Students who learn it combine linear algebra, polynomial approximation, calculus, continuum mechanics and programming—skills that transfer to simulation, data science and numerical software.