Haar measure and integration
Construct translation-invariant volume and compute the left/right discrepancy through an explicitly defined modular function.
On this page
The question for today
What you’ll understand
- State uniqueness and normalization of left Haar measure.
- Translate between right-set translation and the integral modular identity.
- Derive different left and right invariant densities for the positive affine group.
An average over a symmetry group should not depend on where the identity happens to be placed by a group translation. Haar measure provides the corresponding invariant notion of volume. On a Lie group it can be built from a left-invariant top-degree density, but left invariance need not imply right invariance. The modular function measures that discrepancy. Different books place its inverse in different defining formulas, so we will state the set and integral conventions explicitly and check them with a direct coordinate Jacobian calculation.
Translate one volume element everywhere
Choose a nonzero top-degree density at the identity of a finite-dimensional Lie group and transport it by left translations. Integrating that smooth positive density gives a nonzero left-invariant Radon measure, finite on compact sets. Haar's uniqueness theorem says that any two left Haar measures differ by a positive constant factor. The more general existence theorem applies to locally compact Hausdorff groups, but the invariant-density construction is the concrete Lie group version used here.
A compact group has finite positive Haar volume and can therefore be normalized to total mass one. That normalized measure is also right invariant. One way to see this is that a right translate of a left Haar measure is another left Haar measure, and its total mass has not changed. Noncompact groups generally cannot use this probability normalization: a nonzero Haar measure has finite total mass exactly when the group is compact. Uniform group averages and finite probability averages are therefore not interchangeable in noncompact settings.
The modular function detects non-unimodularity
The positive function Delta is a continuous group homomorphism into the multiplicative positive reals. A group is unimodular when it is identically one, meaning that left Haar measure is also right invariant. For a Lie group with the displayed convention, Delta-g is the inverse absolute determinant of Ad-g. Along exp tX this becomes the exponential of minus t times the trace of ad-X. This formula links infinitesimal volume distortion to finite conjugation and gives a convenient independent check on coordinate calculations.
Abelian groups are unimodular because their adjoint action is trivial. Compact groups are unimodular because the image of a compact group under Delta is a compact subgroup of the positive reals, hence trivial. Connected semisimple Lie groups are also unimodular because their infinitesimal adjoint maps have zero trace. None of these facts says that an arbitrary noncompact matrix group has its ordinary Lebesgue measure as Haar measure. A coordinate-dependent density is often essential.
Use a density, not a coordinate rectangle, as the invariant object
For the positive affine group, ordinary parameter area changes under both left and right multiplication. A density proportional to one over the square of the scale corrects left multiplication, while one over the scale corrects right multiplication. The two differ, and the group's modular function records that difference. Translating a coordinate rectangle while keeping its unweighted area fixed is therefore not an invariant-volume test. Both the coordinate Jacobian and the density at the transformed coordinates must be included.
On a compact group, normalized Haar measure allows averaging an inner product or a function without choosing a preferred group element. For a finite-dimensional continuous representation, averaging a positive inner product gives an invariant positive inner product. This is one reason compact-group representation theory is especially well behaved. On a noncompact group that same unweighted integral may diverge, so an argument requiring a normalized invariant average must retain the compactness hypothesis or provide a different construction.
Work it through
Compute left and right volumes for the affine group
For (a,b)(α,β)=(aα,b+aβ), with positive scales, verify left Haar density da db/a², right Haar density da db/a, and Δ(α,β). Check the right translate of 1≤a≤2, 0≤b≤1 by (2,0).
Left multiplication by (α,β) maps (a,b) to (αa,β+αb) and has Jacobian α². Evaluating da db/a² at the new coordinates divides by α²a², so the factors cancel. This proves left invariance of that density.
Right multiplication maps (a,b) to (aα,b+aβ), with Jacobian α. The same left density pulls back to one over α times itself, so Δ(α,β)=1/α. The density da db/a instead has canceling right factors α/(aα), proving right invariance.
The original rectangle has left Haar volume . Its right translate by (2,0) has 2≤a≤4 and the same b interval, giving volume 1/4. This is Delta times the original volume. The corresponding integral of f(xg), rather than a translated set, scales by the reciprocal factor two.
Left and right Haar measures genuinely differ in this group, and the explicitly chosen modular convention agrees with both the Jacobian and the adjoint determinant.
The living notebook
Explore the model
Interactive lab · matrix
Separate coordinate area from weighted group volume
Pause, Reset, and set a=2 and b=0 to inspect the coordinate map diag(2,1), the right multiplication Jacobian for g=(2,0) in the worked affine coordinates. Predict doubled Euclidean area. Compare from Reset with both a=1 and b=0. The group density 1/a² is not drawn: its transformed factor is one quarter, so left Haar volume halves even as coordinate area doubles.
- Area scale · det A
- 1.4
- Singular values · σmax, σmin
- 1.597, 0.876
- Condition number · κ₂(A)
- 1.823
The selected matrix is A = [[a,b],[0,1]]. Playback interpolates geometry between the identity and A; readouts always describe A. Its determinant is a and its singular values come from AᵀA. Positive a makes every supported matrix invertible. This two-dimensional family is not a general eigensolver or a learned network. Parameter changes pause playback and recompute the model at the current elapsed time. Reset restores both parameters and returns time to zero. Playback starts paused. Adjust either parameter, then choose Play or Step.
Look forAn invariant or modular volume calculation must combine a Jacobian with the transformed density; a unit-square area alone measures a different object.
Pause & reason
Check your understanding
Your turn
Try it yourself
On S¹ with normalized Haar measure dθ/(2π), compute the averages of cos θ and cos² θ. Explain which compact-group normalization makes the answers independent of an initial angular shift.
Take a moment to make a prediction or work it out on paper. Open a hint when you need one.
Give me a hint
Integrate over one full period and use cos²θ=(1+cos2θ)/2.
Compare with the solution
The average of cos θ is zero, while the average of cos²θ is one half. The factor 1/(2π) makes the total circle mass one. Replacing θ by θ+φ leaves both full-period integrals unchanged because normalized Haar measure is translation invariant. This compact-group probability average does not imply that a noncompact group has a finite normalized Haar measure.
Keep the idea
Take it with you
- Haar measure is unique up to positive scale, and compact groups admit probability normalization.
- The modular function's inverse placement must be fixed by explicit translation formulas.
- Coordinate Jacobians and invariant densities must be combined before making a volume claim.
Read further
Original sources and deeper explanations to follow at your own pace.
Optional progress, saved in this browser when available. Revisit any lesson at any time.
Lesson not marked complete.