Chapter map

1. 1. Operator
2. 2. Forcing
3. 3. Forcing basis
4. 4. Particular basis
5. 5. Fit
6. 6. Recover
7. 7. Boundary
8. 8. Correction
9. 9. Final
10. 10. RBF extension
11. 11. Condition
12. 12. Conclusion

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

---

1. Operator

This is where modelling choices begin to affect evidence. Set L=d squared/dx squared.

A concrete situation

Linear teaching operator.

What the mathematics does not license

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

Try it, check it, transfer it

Record L. 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. Forcing

The central mathematical idea is concrete. Use f=2.

A concrete situation

Constant source.

What the mathematics does not license

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

Try it, check it, transfer it

Record f. 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. Forcing basis

Here the geometry becomes operational. Choose phi=1.

A concrete situation

Exact constant span.

What the mathematics does not license

One coefficient. The same reasoning helps students move from an exercise to a real dataset.

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.

---

4. Particular basis

This step deserves a careful pause. Choose psi=x squared/2.

A concrete situation

L psi=1.

What the mathematics does not license

Differentiate twice. 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.

---

5. Fit

The useful question is not only how to calculate, but what the calculation assumes. Set alpha=2.

A concrete situation

f=alpha phi.

What the mathematics does not license

Interior match. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record coefficient. 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. Recover

A small example makes the mechanism visible. u_p=alpha psi=x squared.

A concrete situation

L u_p=2.

What the mathematics does not license

Exact teaching stage. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record field. 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. Boundary

This is where modelling choices begin to affect evidence. u_p(0)=0,u_p(1)=1.

A concrete situation

Right boundary missed.

What the mathematics does not license

Particular is not full solution. That distinction matters because a correct formula can still answer the wrong operational question.

Try it, check it, transfer it

Record residual. 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. Correction

The central mathematical idea is concrete. Add h=-x in traditional split.

A concrete situation

Lh=0.

What the mathematics does not license

Demonstration only. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record field. 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. Final

Here the geometry becomes operational. u=x squared-x.

A concrete situation

Both boundaries zero.

What the mathematics does not license

Check L and boundaries. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Record residuals. 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. RBF extension

This step deserves a careful pause. Use multiple phi_j and psi_j.

A concrete situation

Fit nonconstant forcing.

What the mathematics does not license

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

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.

---

11. Condition

The useful question is not only how to calculate, but what the calculation assumes. Inspect operator-image rank.

A concrete situation

Basis pairs can be dependent.

What the mathematics does not license

Use singular values. The habit of checking assumptions is one of the practical benefits of learning mathematics.

Try it, check it, transfer it

Record metric. 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. Conclusion

A small example makes the mechanism visible. State operator, basis pairs and validation.

A concrete situation

Narrow polynomial example.

What the mathematics does not license

Full MAPS may differ. The same reasoning helps students move from an exercise to a real dataset.

Try it, check it, transfer it

Write 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 method of approximate particular solutions begins from the operator equation L u=f. It chooses basis functions for f, pairs them with functions whose image under L is known, and uses collocation to build an approximate solution without requiring a closed-form particular solution for every new forcing term.

For a transparent one-dimensional operator, let L=d squared/dx squared and f(x)=2 on [0,1]. Use the forcing basis phi(x)=1 and the paired particular basis psi(x)=x squared/2, because psi double prime=1.

Approximating f=alpha phi at any interior point gives alpha=2. The corresponding approximate particular field is u_p=alpha psi=x squared, and L u_p=2 exactly for this teaching case.

The field u_p does not satisfy u(0)=u(1)=0. In the traditional decomposition, add the homogeneous correction h=-x, whose second derivative is zero, giving u=x squared-x.

If the operator equation is instead -u double prime=2, the sign changes and the boundary solution becomes u=x(1-x). The sign convention must be fixed before any basis pair is built.

The published MAPS formulation develops collocation machinery so an approximate particular-solution strategy can handle the inhomogeneous problem without separately deriving an analytical homogeneous solution. The small calculation here isolates the operator-pairing mechanism rather than reproducing every formulation detail.

Did You Know? A particular solution only needs to produce the forcing under the differential operator. It does not need to satisfy the homogeneous equation, and it is not unique because any homogeneous function can be added.

With radial basis functions, one may choose phi_j centred at knots and derive or construct psi_j satisfying L psi_j=phi_j. Coefficients fitted to f then transfer directly to the particular field.

The basis pair depends on the operator. A function paired for the Laplacian is not automatically paired for advection–diffusion, variable coefficients or a nonlinear operator.

Shape parameters, knot placement and coordinate scaling affect both forcing approximation and the recovered field. A small forcing residual can still integrate into a noticeable solution error.

Boundary collocation must be included consistently in the complete formulation. If boundary equations and interior equations use very different units, row scaling influences the numerical solve.

Approximate particular solutions can be combined with MFS, Trefftz or boundary-knot ideas. The combination must state which component handles forcing, which handles homogeneous behaviour and where boundary data enter.

Rank deficiency can arise from repeated knots, flat basis functions or operator images that become nearly dependent. Singular values and polynomial reproduction tests expose these problems.

Students can pair simple polynomials under differentiation, fit a nonconstant f at several knots, rebuild u_p and compare L u_p with the target forcing between knots.

Applications include Poisson, diffusion, Helmholtz-type and other linear PDEs when suitable operator-basis pairs are available. Nonlinearity may require iteration or reformulation rather than a single linear coefficient solve.

A responsible MAPS report states L, f, forcing basis, particular basis, knots, scaling, collocation equations, boundary treatment, conditioning, solver, residuals, refinement and validation.

---

Frequently asked questions

What is a particular solution?

Any field whose image under the differential operator equals the forcing term.

Why is it not unique?

A homogeneous solution can be added without changing the forcing.

What does MAPS approximate?

It builds an approximate particular field from collocated operator-basis pairs.

Why use radial basis functions?

They provide flexible translated bases on scattered knots.

Does a forcing fit guarantee boundary accuracy?

No. Boundary equations must be included in the complete formulation.

What must be reported?

Operator, forcing, paired bases, knots, scaling, boundaries, conditioning, solver, convergence and validation.

---

Primary source and fact-check note

C. S. Chen, C. M. Fan and P. H. Wen, “The Method of Approximate Particular Solutions for Solving Certain Partial Differential Equations,” Numerical Methods for Partial Differential Equations 28(2), 2012. 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

• Method of fundamental solutions
• RBF collocation
• Trefftz method
• 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: mechanism, algebra and evidence

Operator audit

Verify L psi_j=phi_j analytically or numerically.

Forcing audit

Measure the approximation of f at and between knots.

Boundary audit

State how boundary equations enter the full formulation.

Rank audit

Inspect scaled matrices, singular values, refinement and validation.

Worked evidence notebook: reconstruct every number

Evidence notebook 1: Operator

Set L=d squared/dx squared. Linear teaching operator. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. State boundary separately. Record L. Change one input, recompute the affected mechanism and identify whether the difference comes from modelling, integration, basis choice, conditioning, boundary data or nonlinear solving.

Evidence notebook 2: Forcing

Use f=2. Constant source. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. Track sign convention. Record f. Change one input, recompute the affected mechanism and identify whether the difference comes from modelling, integration, basis choice, conditioning, boundary data or nonlinear solving.

Evidence notebook 3: Forcing basis

Choose phi=1. Exact constant span. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. One coefficient. Record basis. Change one input, recompute the affected mechanism and identify whether the difference comes from modelling, integration, basis choice, conditioning, boundary data or nonlinear solving.

Evidence notebook 4: Particular basis

Choose psi=x squared/2. L psi=1. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. Differentiate twice. Record check. Change one input, recompute the affected mechanism and identify whether the difference comes from modelling, integration, basis choice, conditioning, boundary data or nonlinear solving.

Evidence notebook 5: Fit

Set alpha=2. f=alpha phi. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. Interior match. Record coefficient. Change one input, recompute the affected mechanism and identify whether the difference comes from modelling, integration, basis choice, conditioning, boundary data or nonlinear solving.

Evidence notebook 6: Recover

u_p=alpha psi=x squared. L u_p=2. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. Exact teaching stage. Record field. Change one input, recompute the affected mechanism and identify whether the difference comes from modelling, integration, basis choice, conditioning, boundary data or nonlinear solving.

Evidence notebook 7: Boundary

u_p(0)=0,u_p(1)=1. Right boundary missed. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. Particular is not full solution. Record residual. Change one input, recompute the affected mechanism and identify whether the difference comes from modelling, integration, basis choice, conditioning, boundary data or nonlinear solving.

Evidence notebook 8: Correction

Add h=-x in traditional split. Lh=0. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. Demonstration only. Record field. Change one input, recompute the affected mechanism and identify whether the difference comes from modelling, integration, basis choice, conditioning, boundary data or nonlinear solving.

Evidence notebook 9: Final

u=x squared-x. Both boundaries zero. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. Check L and boundaries. Record residuals. Change one input, recompute the affected mechanism and identify whether the difference comes from modelling, integration, basis choice, conditioning, boundary data or nonlinear solving.

Evidence notebook 10: RBF extension

Use multiple phi_j and psi_j. Fit nonconstant forcing. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. State knots. Record matrix. Change one input, recompute the affected mechanism and identify whether the difference comes from modelling, integration, basis choice, conditioning, boundary data or nonlinear solving.

Evidence notebook 11: Condition

Inspect operator-image rank. Basis pairs can be dependent. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. Use singular values. Record metric. Change one input, recompute the affected mechanism and identify whether the difference comes from modelling, integration, basis choice, conditioning, boundary data or nonlinear solving.

Evidence notebook 12: Conclusion

State operator, basis pairs and validation. Narrow polynomial example. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. Full MAPS may differ. Write claim. Change one input, recompute the affected mechanism and identify whether the difference comes from modelling, integration, basis choice, conditioning, boundary data or nonlinear solving.

Misconceptions, failure modes and recovery

A particular solution satisfies boundaries

It only matches the forcing unless boundaries are built in. Build a minimal counterexample, restore the missing condition and recalculate before making a broader claim.

Particular solutions are unique

Homogeneous fields can be added. Build a minimal counterexample, restore the missing condition and recalculate before making a broader claim.

One basis pair fits every operator

L determines the pairing. Build a minimal counterexample, restore the missing condition and recalculate before making a broader claim.

Small forcing residual proves field accuracy

Operator inversion can amplify errors. Build a minimal counterexample, restore the missing condition and recalculate before making a broader claim.

Flat RBFs are always accurate

Conditioning can fail. Build a minimal counterexample, restore the missing condition and recalculate before making a broader claim.

Boundary rows need no scaling

Units and magnitude matter. Build a minimal counterexample, restore the missing condition and recalculate before making a broader claim.

MAPS is identical to MFS

They address different operator components. Build a minimal counterexample, restore the missing condition and recalculate before making a broader claim.

One example proves nonlinear applicability

Nonlinear operators need iteration or reformulation. Build a minimal counterexample, restore the missing condition and recalculate before making a broader claim.

A responsible eight-step decision path

1. Define L and f

State variables, units, domain and target claim.

2. Choose forcing basis

Write the mechanism and its exact mathematical promise.

3. Build particular pairs

Keep local computation and full-model assumptions separate.

4. Fit coefficients

Record matrix ordering, residual definitions and tolerances.

5. Assemble boundaries

Measure conditioning, rank and solver convergence.

6. Check rank

Challenge difficult data, geometry and perturbed inputs.

7. Refine knots

Repeat with systematic resolution or parameter changes.

8. Validate field

Compare with independent evidence and state limits.

Learner lab: build transferable understanding

Polynomial pair

Verify psi double prime=phi. Predict first, show the arithmetic, record one diagnostic, compare a baseline and finish with one supported conclusion plus one limit.

Forcing fit

Fit a constant source. Predict first, show the arithmetic, record one diagnostic, compare a baseline and finish with one supported conclusion plus one limit.

Boundary correction

Show why u_p is incomplete. Predict first, show the arithmetic, record one diagnostic, compare a baseline and finish with one supported conclusion plus one limit.

Nonconstant source

Use polynomial or RBF knots. Predict first, show the arithmetic, record one diagnostic, compare a baseline and finish with one supported conclusion plus one limit.

Operator change

Replace L and rebuild pairs. Predict first, show the arithmetic, record one diagnostic, compare a baseline and finish with one supported conclusion plus one limit.

Scaling

Map the interval. Predict first, show the arithmetic, record one diagnostic, compare a baseline and finish with one supported conclusion plus one limit.

Rank

Duplicate a basis centre. Predict first, show the arithmetic, record one diagnostic, compare a baseline and finish with one supported conclusion plus one limit.

Validation

Compare L u with f between knots. Predict first, show the arithmetic, record one diagnostic, compare a baseline and finish with one supported conclusion plus one limit.

Student explanation

Explain operator-paired functions. Predict first, show the arithmetic, record one diagnostic, compare a baseline and finish with one supported conclusion plus one limit.

Parent explanation

Relate matching causes to effects. Predict first, show the arithmetic, record one diagnostic, compare a baseline and finish with one supported conclusion plus one limit.

Deeper transfer: from mathematical structure to trustworthy computation

Operator inversion

Choosing psi_j is an approximate inverse step: the basis is designed so applying L returns a convenient forcing basis. This embeds operator knowledge before coefficients are solved.

Polynomial augmentation

Polynomials can guarantee reproduction of low-order forcing fields and constrain RBF systems. Side conditions must be added consistently.

Variable coefficients

If L changes with position, translated operator pairs may no longer be reusable. Local approximations, iterative linearisation or different bases may be required.

Boundary dominance

For elliptic problems, a small interior forcing error and a small boundary error can have different effects. Report both and compare against an energy or field norm.

Dual reciprocity

Dual reciprocity similarly expands forcing in radial functions and converts domain terms into boundary-friendly particular solutions. MAPS belongs in this broader operator-splitting landscape.

Shape parameters

Very flat RBFs can improve approximation in exact arithmetic while producing ill-conditioned matrices. Stable algorithms and precision studies are needed.

Manufactured solutions

Choose u_exact first, calculate f=L u_exact and boundary data, then test the complete method. This exposes sign, operator and boundary mistakes.

Career connection

MAPS combines differential operators, basis design, collocation and numerical linear algebra for PDE software.

Student and parent guidance: learning the method without losing the meaning

A student’s first route

Pair simple polynomials under differentiation before moving to radial functions and scattered knots.

A parent’s useful question

Ask the learner to distinguish the forcing equation from boundary conditions and the full solution.

A small coding exercise

Fit f at knots, recover u_p from paired bases and test L u_p between knots.

An engineering exercise

Use a manufactured Poisson solution and compare polynomial, Gaussian and polyharmonic forcing bases.

A verification habit

Track forcing fit, operator residual, boundary residual, rank, conditioning and refinement separately.

A career pathway

PDEs, basis design and numerical linear algebra support simulation, modelling and scientific software.

Scope, evidence and honest limits

What the worked example proves

It verifies the polynomial operator pair L(x squared/2)=1 and a traditional homogeneous correction.

What a software result must add

A MAPS computation adds the actual collocation formulation, knots, radial bases, boundary equations, conditioning and refinement.

What to compare

Use manufactured solutions and measure forcing, boundary and field errors on points not used for collocation.

What to preserve

Save operator definitions, basis pairs, knots, matrices, coefficients, residuals and convergence tables.

A final communication check

Explain the method to a student and to a parent while preserving the mechanism, arithmetic, evidence and limit.

Field notes: eight deeper checks before trusting the result

Basis-pair ledger

For every phi_j, record psi_j and a numerical test that L psi_j equals phi_j over the domain.

Off-knot checks

Collocation can hide oscillation between knots. Evaluate forcing and field errors on an independent grid.

Row scaling

Balance interior operator rows and boundary rows using physical nondimensionalisation or a declared numerical scaling.

Shape sweep

Vary RBF shape parameters and precision together; report error against condition number.

Stopping rule

Require stable field values, forcing residuals, boundary residuals and refinement rates.

Failure recovery

Change knots, add polynomial terms, use stable RBF algorithms or choose a basis better matched to L.

Portfolio evidence

Publish operator tests, matrices, singular values and manufactured-solution errors.

Transfer

MAPS teaches students to design functions by what an operator does to them, a powerful inverse way to think about calculus.

Further practice: six transferable checks

Green identity perspective

Traditional particular-plus-homogeneous decompositions mirror the linear superposition behind Green-function methods. MAPS changes how the particular field is constructed while preserving the need to respect the full operator and boundaries.

Derivative accuracy

A basis can fit f well in value yet produce an inaccurate u when repeated integration magnifies low-frequency errors. Measure the final field and relevant derivatives, not only the source fit.

Boundary knot balance

Too many interior knots with too few boundary equations can bias the least-squares system, while the reverse can underresolve forcing. State counts, weights and sampling geometry.

Noisy forcing

If f comes from measurements, exact interpolation may reproduce noise. Regularisation or smoothing changes the inferred field and requires an uncertainty or sensitivity analysis.

Solver choice

Dense direct factorisation is transparent for small systems. Larger or ill-conditioned problems may need pivoting, truncated decompositions, preconditioning or stable RBF algorithms, all reported with tolerances.

Learning transfer

Operator-pair design teaches reverse reasoning: choose a simple effect under differentiation, then build the cause whose derivative produces it.