Chapter map

1. 1. Operator
2. 2. Knots
3. 3. Basis
4. 4. Matrix
5. 5. Data
6. 6. Determinant
7. 7. Solve
8. 8. Midpoint
9. 9. Exact check
10. 10. Neumann
11. 11. Forcing
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

A small example makes the mechanism visible. Set u second-u=0.

A concrete situation

Modified Helmholtz analogue.

What the mathematics does not license

Homogeneous teaching problem. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Record PDE. 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. Knots

This is where modelling choices begin to affect evidence. Use x0=0,x1=1.

A concrete situation

Boundary centres.

What the mathematics does not license

No fictitious curve. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Record points. 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. Basis

The central mathematical idea is concrete. Use cosh|x-xj|.

A concrete situation

Nonsingular general solutions.

What the mathematics does not license

One-dimensional analogue. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Record functions. 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. Matrix

Here the geometry becomes operational. Assemble [[1,C],[C,1]].

A concrete situation

C=cosh1.

What the mathematics does not license

Keep ordering. Evidence becomes stronger when the limitation is tested instead of hidden.

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.

---

5. Data

This step deserves a careful pause. Use [1,3].

A concrete situation

Dirichlet values.

What the mathematics does not license

State units. Keeping this boundary clear makes the result easier to audit and transfer.

Try it, check it, transfer it

Record vector. 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

The useful question is not only how to calculate, but what the calculation assumes. Compute 1-C squared.

A concrete situation

About -1.3811.

What the mathematics does not license

Nonzero here. A trustworthy workflow records the choice before seeing the final test result.

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.

---

7. Solve

A small example makes the mechanism visible. Get c0 about 2.6277,c1 about -1.0549.

A concrete situation

Signed coefficients.

What the mathematics does not license

Check residual. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Record vector. 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. Midpoint

This is where modelling choices begin to affect evidence. Use cosh(.5).

A concrete situation

u(.5) about 1.7736.

What the mathematics does not license

Both knots contribute. Keeping this boundary clear makes the result easier to audit and transfer.

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. Exact check

The central mathematical idea is concrete. Compare cosh-sinh form.

A concrete situation

Agreement to rounding.

What the mathematics does not license

Teaching span is complete. A trustworthy workflow records the choice before seeing the final test result.

Try it, check it, transfer it

Record error. 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. Neumann

Here the geometry becomes operational. Differentiate basis with direction.

A concrete situation

Normals change row.

What the mathematics does not license

Do not reuse value row. Evidence becomes stronger when the limitation is tested instead of hidden.

Try it, check it, transfer it

Record formula. 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. Forcing

This step deserves a careful pause. Add DRM particular part when needed.

A concrete situation

Homogeneous basis alone is insufficient.

What the mathematics does not license

State split. Keeping this boundary clear makes the result easier to audit and transfer.

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.

---

12. Conclusion

The useful question is not only how to calculate, but what the calculation assumes. State basis, knots and validation.

A concrete situation

Exact 1D example.

What the mathematics does not license

General BKM needs completeness checks. A trustworthy workflow records the choice before seeing the final test result.

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 boundary knot method is a boundary-type radial-basis collocation approach that uses nonsingular general solutions for the homogeneous part and dual-reciprocity or related particular-solution machinery for inhomogeneous terms.

To see the nonsingular boundary-knot idea in one dimension, consider u double prime-u=0 on [0,1] with u(0)=1 and u(1)=3. A nonsingular general solution centred at a knot x_j is cosh|x-x_j|.

Place boundary knots at x0=0 and x1=1 and write u_h=c0 cosh|x|+c1 cosh|x-1|.

Let C=cosh(1), approximately 1.543081. Collocating the two boundary values gives the dense system [[1,C],[C,1]][c0,c1]^T=[1,3]^T.

The determinant is 1-C squared, approximately -1.381098. Solving gives c0 approximately 2.6277 and c1 approximately -1.0549.

At x=0.5 both distances are 0.5, and cosh(0.5) is approximately 1.127626. The midpoint value is (c0+c1)cosh(0.5), approximately 1.7736.

The exact form u=A cosh x+B sinh x has A=1 and B=(3-cosh1)/sinh1. It gives the same midpoint to rounding, so this tiny boundary-knot span is exact for the one-dimensional homogeneous solution space.

Did You Know? Unlike a singular fundamental solution, a nonsingular general solution can place its centre directly on the physical boundary. This removes the fictitious source boundary used by MFS for eligible operators.

For inhomogeneous PDEs, BKM commonly uses dual reciprocity: radial basis functions approximate the forcing and paired particular solutions convert that approximation into a field contribution.

Nonsingular general solutions are operator specific. For two-dimensional Helmholtz problems they involve Bessel functions; for modified Helmholtz problems they involve modified Bessel functions. The one-dimensional cosh example is a teaching analogue.

Completeness is a substantive question because using only the nonsingular part of a complete fundamental solution may restrict the approximation space. Numerical success on one geometry does not settle the theory for every operator and domain.

Boundary normals matter for Neumann and Robin conditions. A derivative row changes with knot-to-point direction and outward normal, while mixed conditions can alter symmetry.

The boundary system is dense because every knot influences every collocation point. Symmetry may occur under specific choices, but it should be verified from the saved matrix rather than assumed from the method label.

Students can rebuild the 2 by 2 hyperbolic-cosine system, compare it with the equivalent cosh–sinh basis and perturb a knot to see how conditioning changes.

Applications reported for BKM include Helmholtz, diffusion and convection–diffusion problems on complex geometry. A method that avoids meshes still needs boundary representation, knot-quality checks and independent validation.

A responsible BKM report states the PDE, homogeneous and inhomogeneous split, nonsingular general solution, RBF and particular-solution construction, boundary knots, normals, scaling, solver, conditioning, convergence and validation.

---

Frequently asked questions

What is the boundary knot method?

A boundary-type RBF collocation method using nonsingular general solutions for the homogeneous field.

How does it differ from MFS?

Eligible nonsingular bases can be centred on the physical boundary rather than a fictitious source boundary.

How are forcing terms handled?

Common formulations use dual reciprocity and particular-solution bases.

Is BKM always boundary only?

That depends on the operator, forcing treatment and formulation; the actual knots and equations must be stated.

Why is completeness important?

The nonsingular part alone may not span every required solution behaviour.

What must be reported?

PDE, split, general solution, RBFs, knots, normals, conditioning, convergence and validation.

---

Primary source and fact-check note

W. Chen and M. Tanaka, “A Meshless, Integration-Free, and Boundary-Only RBF Technique,” Computers & Mathematics with Applications 43, 2002. 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
• Trefftz method
• Method of approximate particular solutions source
• Computational mathematics hub
• 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: mechanism, algebra and evidence

Basis audit

Verify every nonsingular general solution satisfies the homogeneous operator.

Knot audit

Record boundary positions, normals and ordering.

Forcing audit

Separate the homogeneous boundary system from the particular contribution.

Completeness audit

Check rank, refinement and comparison with an independent solution.

Worked evidence notebook: reconstruct every number

Evidence notebook 1: Operator

Set u second-u=0. Modified Helmholtz analogue. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. Homogeneous teaching problem. Record PDE. 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: Knots

Use x0=0,x1=1. Boundary centres. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. No fictitious curve. Record points. 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: Basis

Use cosh|x-xj|. Nonsingular general solutions. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. One-dimensional analogue. Record functions. 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: Matrix

Assemble [[1,C],[C,1]]. C=cosh1. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. Keep ordering. 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 5: Data

Use [1,3]. Dirichlet values. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. State units. Record vector. 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: Determinant

Compute 1-C squared. About -1.3811. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. Nonzero here. Record value. 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: Solve

Get c0 about 2.6277,c1 about -1.0549. Signed coefficients. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. Check residual. Record vector. 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: Midpoint

Use cosh(.5). u(.5) about 1.7736. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. Both knots contribute. Record value. 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: Exact check

Compare cosh-sinh form. Agreement to rounding. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. Teaching span is complete. Record error. 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: Neumann

Differentiate basis with direction. Normals change row. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. Do not reuse value row. Record formula. 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: Forcing

Add DRM particular part when needed. Homogeneous basis alone is insufficient. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. State split. 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 12: Conclusion

State basis, knots and validation. Exact 1D example. Keep variables, units, coordinates, matrix entries, signs, tolerances and rounding visible. General BKM needs completeness checks. 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

Every PDE has a useful nonsingular basis

Operator-specific availability and completeness matter. Build a minimal counterexample, restore the missing condition and recalculate before making a broader claim.

Boundary only means geometry free

Boundary knots and normals still define geometry. Build a minimal counterexample, restore the missing condition and recalculate before making a broader claim.

No fictitious boundary means no parameters

Knot placement, RBFs and scaling remain. Build a minimal counterexample, restore the missing condition and recalculate before making a broader claim.

The system is always symmetric

Symmetry depends on formulation and conditions. Build a minimal counterexample, restore the missing condition and recalculate before making a broader claim.

Nonsingular means well conditioned

Dense columns may still be dependent. Build a minimal counterexample, restore the missing condition and recalculate before making a broader claim.

Forcing disappears

A particular-solution treatment remains. Build a minimal counterexample, restore the missing condition and recalculate before making a broader claim.

One-dimensional cosh proves 2D completeness

It is only a teaching analogue. Build a minimal counterexample, restore the missing condition and recalculate before making a broader claim.

Mesh-free means verification-free

Refinement and validation remain essential. Build a minimal counterexample, restore the missing condition and recalculate before making a broader claim.

A responsible eight-step decision path

1. Define PDE split

State variables, units, domain and target claim.

2. Choose general solution

Write the mechanism and its exact mathematical promise.

3. Place boundary knots

Keep local computation and full-model assumptions separate.

4. Build value or flux rows

Record matrix ordering, residual definitions and tolerances.

5. Add forcing part

Measure conditioning, rank and solver convergence.

6. Solve with diagnostics

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

Cosh basis

Verify u second-u=0. Predict first, show the arithmetic, record one diagnostic, compare a baseline and finish with one supported conclusion plus one limit.

Boundary matrix

Rebuild the 2 by 2 system. Predict first, show the arithmetic, record one diagnostic, compare a baseline and finish with one supported conclusion plus one limit.

Midpoint

Recover 1.7736. Predict first, show the arithmetic, record one diagnostic, compare a baseline and finish with one supported conclusion plus one limit.

Exact form

Compare cosh-sinh coefficients. Predict first, show the arithmetic, record one diagnostic, compare a baseline and finish with one supported conclusion plus one limit.

Knot move

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

Neumann row

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

Forcing

Add a simple particular field. Predict first, show the arithmetic, record one diagnostic, compare a baseline and finish with one supported conclusion plus one limit.

Conditioning

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

Student explanation

Explain boundary-centred solutions. Predict first, show the arithmetic, record one diagnostic, compare a baseline and finish with one supported conclusion plus one limit.

Parent explanation

Relate edge anchors to an interior field. 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

General solutions

A general solution satisfies the homogeneous operator without a singularity at its centre. Bessel, modified Bessel or exponential forms appear depending on dimension and PDE.

Dual reciprocity

Approximate f by RBFs, map each basis to a particular solution and subtract its boundary trace before solving the homogeneous knot system.

Mixed conditions

Dirichlet rows use basis values; Neumann rows use normal derivatives. Symmetry can be lost when row and centre operators differ.

Geometry representation

Boundary-only knots reduce dimensional sampling but still need accurate curved surfaces, normals and corner treatment.

Completeness

A nonsingular subset may omit solution modes needed by a domain. Enrichment, multiple centres or alternative formulations may be necessary.

Eigenproblems

Boundary-knot matrices depending on a spectral parameter can be used to search for eigenvalues, with determinant or singular-value diagnostics.

Fast algebra

Dense RBF interactions invite hierarchical matrices or fast kernel evaluation for large boundaries, but preconditioning remains operator dependent.

Career connection

BKM connects special functions, radial bases, boundary geometry and dense linear algebra in computational physics.

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

A student’s first route

Verify cosh solves the operator, build the 2 by 2 boundary matrix and compare with the ordinary exact basis.

A parent’s useful question

Ask why nonsingular centres may sit on the boundary while MFS sources are kept outside.

A small coding exercise

Move boundary knots, rebuild coefficients and plot midpoint error and condition number.

An engineering exercise

Solve a known Helmholtz or modified-Helmholtz problem on a simple 2D boundary and refine knots.

A verification habit

Track homogeneous residual, boundary residual, forcing approximation, rank, knot refinement and independent error.

A career pathway

Special functions, boundary geometry and RBF algebra support acoustics, diffusion and computational physics.

Scope, evidence and honest limits

What the worked example proves

It verifies a two-knot nonsingular cosh span for one one-dimensional homogeneous problem.

What a software result must add

A BKM computation adds the correct dimensional general solution, forcing treatment, curved geometry, normals and completeness evidence.

What to compare

Use an analytical or independently refined field while varying knot density, placement and arithmetic precision.

What to preserve

Save knots, normals, basis formulas, matrices, singular values, coefficients, residuals and refinement data.

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

Operator table

List the PDE, dimension and nonsingular general solution before assembling any row.

Knot geometry

Plot knot spacing, normals and curvature; a count alone does not describe resolution.

Boundary tests

Evaluate value and flux residuals between knots, not only at collocation locations.

Completeness challenge

Test solutions with different symmetry, corner behaviour and modes to expose missing basis content.

Stopping rule

Require stable errors under knot refinement, offset-free geometry changes and increased precision.

Failure recovery

Enrich the basis, split the domain, improve normals or use a method with a more complete representation.

Portfolio evidence

Save geometry, special-function evaluations, matrices, spectra and comparisons.

Transfer

BKM shows how knowing exact local physics can reduce dimensional work while leaving geometry and algebra fully accountable.

Further practice: six transferable checks

Special-function evaluation

Bessel and modified-Bessel functions need stable numerical libraries, especially for large arguments or high orders. Overflow, underflow and cancellation can appear before the linear solve begins.

Corner normals

At a geometric corner the outward normal is not unique. One-sided boundary segments, averaged normals or a weak treatment are different choices and should not be hidden behind a single knot.

Knot refinement

Refine according to boundary length, curvature, wavelength and solution features. Doubling knot count uniformly is a useful baseline, not the only defensible strategy.

Exterior problems

Radiation conditions and unbounded domains require bases that represent outgoing behaviour. A nonsingular interior general solution does not automatically satisfy an exterior radiation condition.

Comparison discipline

Compare BKM and MFS with the same geometry, boundary data, arithmetic precision and validation points. Otherwise differences may come from source placement or sampling rather than the core formulation.

Learning transfer

The method shows that exact operator knowledge can be packaged into basis functions, while geometry, data and matrix evidence still decide whether the computation is trustworthy.

A reproducible boundary study

Solve one manufactured problem on successively refined knot sets. For each set, report spacing, curvature coverage, boundary value and flux residuals, condition number, coefficient norm and independent interior error. Repeat with a second solution mode to challenge completeness. This table distinguishes geometric resolution, algebraic sensitivity and basis limitations more clearly than one attractive contour plot.

Naming the formulation

Boundary-knot publications include variations in forcing approximation, basis selection and symmetry. Name the actual equations and basis functions rather than assuming the label alone tells a reader how the matrix was built.