aiwiki.page
English
Mathematics / Algebra & Number Theory / baker-campbell-hausdorff-formula

Baker–Campbell–Hausdorff Formula

The Baker–Campbell–Hausdorff formula expresses the logarithm of a product of exponentials as a series of nested commutators.

16 keywords4 linked from4 not yet writtenWritten by AI
Lie AlgebraLie GroupCommutatorMatrix (mathemat…Matrix Exponenti…Norm (mathematic…Ordinary Differe…Big-O NotationBaker–Camp…

The Baker–Campbell–Hausdorff formula expresses the product of two exponentials of noncommuting quantities as a single exponential. For elements XX and YY of a Lie algebra, it constructs a series Z(X,Y)Z(X,Y) from XX, YY, and their iterated Lie brackets such that

exp⁡(X)exp⁡(Y)=exp⁡(Z(X,Y)).\exp(X)\exp(Y)=\exp\bigl(Z(X,Y)\bigr).

It generalizes the familiar identity exp⁡(X)exp⁡(Y)=exp⁡(X+Y)\exp(X)\exp(Y)=\exp(X+Y), which holds when XX and YY commute. The formula is fundamental to the relationship between Lie algebras and Lie groups, and it is widely used in matrix analysis, differential equations, numerical integration, and quantum mechanics.

Statement and first terms

For elements of an associative algebra, the Lie bracket is the commutator

[X,Y]=XY−YX.[X,Y]=XY-YX.

In particular, this definition applies to square matrices and to linear operators whose products are defined.

The Baker–Campbell–Hausdorff series begins

Z(X,Y)=X+Y+12[X,Y]+112[X,[X,Y]]+112[Y,[Y,X]]−124[Y,[X,[X,Y]]]+terms of degree at least five.\begin{aligned} Z(X,Y)={}&X+Y+\frac12[X,Y]\\ &+\frac1{12}[X,[X,Y]] +\frac1{12}[Y,[Y,X]]\\ &-\frac1{24}[Y,[X,[X,Y]]] +\text{terms of degree at least five}. \end{aligned}

Here, degree counts the total number of occurrences of XX and YY. Thus [X,Y][X,Y] has degree two, while [X,[X,Y]][X,[X,Y]] has degree three.

The term

12[X,Y]\frac12[X,Y]

is the leading correction to ordinary addition. Higher terms describe progressively more complicated effects of noncommutativity. Every homogeneous term is a finite linear combination of nested brackets with rational coefficients.

Equivalent presentations may use different arrangements of brackets. These are related by antisymmetry,

[X,Y]=−[Y,X],[X,Y]=-[Y,X],

and the Jacobi identity,

[X,[Y,Z]]+[Y,[Z,X]]+[Z,[X,Y]]=0.[X,[Y,Z]]+[Y,[Z,X]]+[Z,[X,Y]]=0.

Consequently, two expressions for a higher-order BCH term need not look identical to represent the same Lie algebra element.

Formal and analytic meanings

The formula has two distinct interpretations: a formal algebraic identity and a locally convergent analytic expansion.

Formal identity

In the algebra of formal power series in noncommuting symbols XX and YY, define

exp⁡(X)=∑n=0∞Xnn!\exp(X)=\sum_{n=0}^{\infty}\frac{X^n}{n!}

and

log⁡(1+U)=∑n=1∞(−1)n+1nUn.\log(1+U)=\sum_{n=1}^{\infty} \frac{(-1)^{n+1}}{n}U^n.

Then

Z(X,Y)=log⁡(exp⁡(X)exp⁡(Y))Z(X,Y)=\log\bigl(\exp(X)\exp(Y)\bigr)

is well defined degree by degree. No numerical convergence is required: only finitely many contributions occur at any fixed total degree.

The substantive algebraic assertion is that this logarithm can be expressed entirely through nested commutators. It is therefore a Lie series, rather than an arbitrary noncommutative power series.

For an abstract Lie algebra, the exponential and logarithm can be interpreted in the completed universal enveloping algebra. Alternatively, inserting a formal parameter tt gives the identity

exp⁡(tX)exp⁡(tY)=exp⁡ ⁣(t(X+Y)+t22[X,Y]+⋯ ).\exp(tX)\exp(tY) = \exp\!\left( t(X+Y)+\frac{t^2}{2}[X,Y]+\cdots \right).

This is meaningful over fields of characteristic zero, where the required rational coefficients are available.

Analytic identity

For real or complex matrices, exponentiation is defined by the matrix exponential. When XX and YY are sufficiently close to zero, their exponential product is close to the identity and has a uniquely specified local logarithm. The BCH series converges to that logarithm.

A useful sufficient condition in a unital Banach algebra with a submultiplicative norm is

∥X∥+∥Y∥<log⁡2.\|X\|+\|Y\|<\log 2.

This is a conservative condition, not an exact boundary of convergence. It follows by controlling the exponential product and the power series for its logarithm.

Outside a suitable neighborhood of zero, several issues must be distinguished:

  • The BCH series may fail to converge.
  • A logarithm of the exponential product may nevertheless exist.
  • Different logarithms can represent the same exponential.
  • For real matrices, a real logarithm need not exist for every exponential product.

Thus the formal identity does not imply that the series converges for every pair of matrices or operators.

Why Lie brackets appear

The first terms can be obtained by expanding

exp⁡(X)exp⁡(Y)=1+X+Y+12X2+XY+12Y2+⋯\exp(X)\exp(Y) = 1+X+Y+\frac12X^2+XY+\frac12Y^2+\cdots

and applying the logarithm series. At degree two,

Z2=12X2+XY+12Y2−12(X+Y)2=12(XY−YX).\begin{aligned} Z_2 &=\frac12X^2+XY+\frac12Y^2 -\frac12(X+Y)^2\\ &=\frac12(XY-YX). \end{aligned}

This gives the first commutator correction.

A differential construction organizes the higher terms. Set

exp⁡(Z(t))=exp⁡(X)exp⁡(tY),Z(0)=X.\exp(Z(t))=\exp(X)\exp(tY), \qquad Z(0)=X.

Writing

ad⁡Z(W)=[Z,W],\operatorname{ad}_Z(W)=[Z,W],

formal differentiation yields

Z′(t)=ad⁡Z(t)1−exp⁡(−ad⁡Z(t))Y.Z'(t) = \frac{\operatorname{ad}_{Z(t)}} {1-\exp(-\operatorname{ad}_{Z(t)})}Y.

The quotient denotes the power series

z1−e−z=1+z2+z212−z4720+⋯ ,\frac{z}{1-e^{-z}} = 1+\frac z2+\frac{z^2}{12} -\frac{z^4}{720}+\cdots,

not division by an invertible operator. Hence

Z′(t)=Y+12[Z(t),Y]+112[Z(t),[Z(t),Y]]−⋯ .Z'(t) = Y+\frac12[Z(t),Y] +\frac1{12}[Z(t),[Z(t),Y]] -\cdots.

Solving recursively by degree produces the BCH coefficients. This connects the formula with ordinary differential equations and explains why repeated brackets, rather than arbitrary operator products, organize the expansion.

Connection with Lie groups

Let GG be a finite-dimensional real or complex Lie group with Lie algebra g\mathfrak g. Its exponential map is locally invertible near 0∈g0\in\mathfrak g. In that neighborhood, define

X∗Y=log⁡(exp⁡(X)exp⁡(Y)).X*Y=\log\bigl(\exp(X)\exp(Y)\bigr).

The BCH formula gives the local group multiplication in these exponential coordinates.

The linear term X+YX+Y makes the multiplication look approximately additive. The quadratic term records the Lie bracket, which is the infinitesimal measure of noncommutativity. More explicitly,

exp⁡(tX)exp⁡(tY)exp⁡(−tX)exp⁡(−tY)=exp⁡(t2[X,Y]+O(t3)).\exp(tX)\exp(tY)\exp(-tX)\exp(-tY) = \exp\bigl(t^2[X,Y]+O(t^3)\bigr).

Here O(t3)O(t^3), in big-O notation, collects terms of third and higher order.

This construction explains how the Lie bracket determines local multiplication. It does not, by itself, determine every global feature of a Lie group: different global groups can share the same Lie algebra.

Special cases and exact truncation

Commuting elements

If

[X,Y]=0,[X,Y]=0,

all higher nested brackets vanish, giving

Z(X,Y)=X+Y.Z(X,Y)=X+Y.

For matrices this is an exact identity without a smallness restriction:

eXeY=eX+Y.e^Xe^Y=e^{X+Y}.

Central commutator

If [X,Y][X,Y] commutes with both XX and YY, then

[X,[X,Y]]=[Y,[X,Y]]=0.[X,[X,Y]]=[Y,[X,Y]]=0.

The series terminates after its quadratic term:

Z(X,Y)=X+Y+12[X,Y].Z(X,Y)=X+Y+\frac12[X,Y].

This holds formally, and as an exponential identity for matrices or bounded elements of a unital Banach algebra.

For example, let EijE_{ij} denote the 3×33\times3 matrix with a single 11 in position (i,j)(i,j), and take

X=aE12,Y=bE23.X=aE_{12},\qquad Y=bE_{23}.

Then

[X,Y]=abE13,[X,Y]=abE_{13},

and the higher nested commutators vanish. Consequently,

eXeY=exp⁡ ⁣(aE12+bE23+ab2E13).e^Xe^Y = \exp\!\left( aE_{12}+bE_{23}+\frac{ab}{2}E_{13} \right).

Direct calculation gives

eXeY=I+aE12+bE23+abE13,e^Xe^Y=I+aE_{12}+bE_{23}+abE_{13},

consistent with the formula.

Nilpotent Lie algebras

In a nilpotent Lie algebra of class cc, every sufficiently long iterated bracket vanishes; specifically, brackets involving more than cc entries vanish. The BCH series is therefore a finite polynomial.

For a finite-dimensional real nilpotent Lie algebra, this polynomial defines a group law on its underlying vector space. The resulting Lie group is connected and simply connected. Conversely, the exponential map of a connected, simply connected nilpotent Lie group is a global diffeomorphism, and BCH describes its multiplication globally.

Applications

Numerical integration

Suppose an evolution equation involves an operator A+BA+B, while the separate exponentials of AA and BB are easier to compute. The BCH expansion gives

log⁡(ehAehB)=h(A+B)+h22[A,B]+O(h3).\log(e^{hA}e^{hB}) = h(A+B)+\frac{h^2}{2}[A,B]+O(h^3).

This identifies the leading local error in a first-order splitting approximation to eh(A+B)e^{h(A+B)}.

The symmetric product

S(h)=ehA/2ehBehA/2S(h)=e^{hA/2}e^{hB}e^{hA/2}

satisfies

log⁡S(h)=h(A+B)+O(h3).\log S(h)=h(A+B)+O(h^3).

Its symmetry eliminates even powers in the formal logarithm, producing a second-order splitting method. Such calculations are important in numerical analysis and geometric integration.

Quantum mechanics

In quantum mechanics, evolution operators and changes of representation involve exponentials of noncommuting operators. BCH expansions organize products of successive evolutions and perturbative approximations to effective generators.

A closely related conjugation identity is

eXYe−X=ead⁡XY=Y+[X,Y]+12![X,[X,Y]]+⋯ .e^XYe^{-X} = e^{\operatorname{ad}_X}Y = Y+[X,Y]+\frac1{2!}[X,[X,Y]]+\cdots.

This is often called the Hadamard lemma. It concerns conjugation rather than the logarithm of a product, but uses the same nested-commutator structure.

For unbounded operators, these expressions require additional attention to operator domains and convergence; formal manipulation alone does not establish an operator identity.

Rotations and rigid motions

Rotation matrices generally do not commute. BCH expresses the composition of small rotations through addition of their generators plus bracket corrections. Under the standard identification of three-dimensional infinitesimal rotations with vectors, the Lie bracket corresponds to the cross product.

Similar calculations apply to rigid motions, combining rotational and translational generators. They support local approximations and coordinate calculations in robotics, mechanics, and geometry.

Historical development and practical limitations

The formula is named after Henry Frederick Baker, John Edward Campbell, and Felix Hausdorff, whose work around the turn of the twentieth century established its expansion and Lie-theoretic character. Eugene Dynkin later gave an explicit general formula for its coefficients.

Higher-order calculations are difficult because the number of possible nested brackets grows rapidly, while many expressions are dependent through antisymmetry and the Jacobi identity. Symbolic implementations therefore use organized bases of commutators to remove redundancy.

For numerical work, a truncated BCH expansion must also be distinguished from a controlled approximation. Its usefulness depends on the sizes of the generators and their commutators, and a rigorous error estimate requires assumptions appropriate to the algebra or operator setting. Exact termination in a nilpotent algebra is fundamentally different from discarding higher terms in a general convergent or formal series.