Lecture II: Hamiltonian formulation of general

relativity
(Courses in canonical gravity)

Yaser Tavakoli

December 16, 2014

1 Space-time foliation
The Hamiltonian formulation of ordinary mechanics is given in terms of a set
of canonical variables q and p at a given instant of time t. In field theory,
however, one is dealing with fields rather than a mechanical system then
the canonical variables ϕ(x) are functions of position, and their canonical
momenta are πϕ (x), both given at an instant of time. General relativity
treats space and time on the same footing, that is not what is done in
Hamiltonian formulations. Therefore, in order to discuss general relativity in
a Hamiltonian fashion, one needs to break that equal footing. This requires a
space-time splitting, since only time derivatives are transformed to momenta
but not space derivatives.
We assume a foliation of space-time in terms of space-like three dimen-
sional surfaces S of space-time manifold M . Thus, we consider the case
of a Lorentzian manifold M diffeomorphic to R × S, where the manifold S
represents ‘space’, and t ∈ R represents ‘time’. It should be noted that, the
particular slicing of space-time into ‘instants of time’ is an arbitrary choice,
rather than something intrinsic to the world. In other words, there are lots
of way to pick a diffeomorphism

ϕ : M −→ R × S.

These give different ways to define a time coordinate τ on the space-time

manifold M , namely the pull-back by ϕ of the standard time coordinate t
on R × S:
τ = ϕ∗ t.
For simplicity, we assume a submanifold Σ ⊂ M is a slice of M if it equals
{τ = const} for some time coordinate τ .

1
Canonical General Relativity 2

2 Geometry of hypersurfaces
Let us consider a surface Σ ≈ Σt0 : t0 = const in a foliated space-time
manifold R×Σ. This can be considered as a constraint surface characterized
by Ct0 = t − t0 = 0. Notice that, the geometry of the constraint surface Σt0
in a space-time is governed by a Riemannian geometry with the metric gab
(with the inverse g ab ) rather than a Poisson or symplectic one on a phase
space.
In analogy to the Poisson geometry, let us associate to the constraint
Ct0 a (Hamiltonian-like) vector field as g ♯ dCt0 on the Riemannian geometry
(which is given by a metric tensor g ab rather than a Poisson tensor P ij ):

g ♯ dCt0 = g ab ∂b Ct0 = g ab ∂b t = g ♯ dt . (2.1)

This shows that, in the herein Riemannian geometry, the vector field X a =
g ab ∂b t is normal1 to the constraint surface Σt0 ; this is opposite to the case
happens in the Poisson geometry where the Hamiltonian vector field of a
single, necessarily first class constraint must be tangent to the constraint
surface. This is, indeed, because of the antisymmetric feature of the Poisson
tensor that makes the Hamiltonian vector field of a single constraint C
tangent to the constraint surface: XC C = P ij ∂i C∂j C = 0.
Using the definition of the normal vector X a we can determine the nor-
malized (time-like) normal vector to the surface as
Xa
na = √ , (2.2)
−gbc X b X c

such that gab na nb = −1. Furthermore, for any vector field sa tangent to Σt0
we have gab sa nb = 0.
The tangent space in T M at each point of Σt can be decomposed to
a ‘spatial tangent space’ spanned by vectors tangent to Σt , and a ‘normal
space’ spanned by the unique unit future-pointing vector field na normal to
Σt . For example, given a vector field Z a ∈ Tp M at any point p ∈ Σt , we
can decompose it into a component tangent to Σt and a normal component
proportional to nb :

Z a = −gab Z a nb + (Z a + gab Z a nb ) . (2.3)

| {z } | {z }
⊥ ∥

Each spatial slice Σt is equipped with its own Riemannian structure. The
induced metric hab on Σt can be uniquely determined by using the two
conditions that

hab na = 0 , and hab sa = gab sa , (2.4)

1
Given any surface C̃(xb ) = 0, the normal vector to C̃ is determined as dC̃ =
(∂ C̃/∂xa )∂a =: Y a ∂a . Thus, Y a = g ab ∂b C̃ is the components of the normal vector.
Canonical General Relativity 3

for any vector sa tangent to Σt . So that, the induced metric hab reads
hab = gab + na nb (2.5)
Interestingly, in comparison to the Poisson geometry, the induced metric
hab is analogous to the Dirac bracket, which subtracts from off the Poisson
structure any contribution from the flow of the constraints transversal to
the constraint surface. The inverse of the induced metric hab can be defined
as hab = g ab + na nb .
In order to study the dynamics of the canonical formulation, we con-
sider an interpretation of the induced metric hab as a time-dependent 3-
dimensional tensor field on the family of manifolds Σt . Thus, the time-
dependent fields hab will play a crucial role as the configuration variables
of canonical gravity. In this way, it makes sense to define time derivatives
of the induced metric or any other fields. Let us introduce a time-evolution
vector field t = ta ∇a to define the direction of time derivatives, such that
ta is normalized: ta ∇a t = 1. By introducing the shift vector N a := hab tb ,
and the laps function N := −nb tb , the time-evolution vector field ta can be
decomposed to the spatial and normal parts as
ta = N na + N a . (2.6)
Using this relation, we can write the inverse space-time metric as
1 a
g ab = hab − na nb = hab − (t − N a )(tb − N b ). (2.7)
N2
By inverting this matrix and writing it in coordinate basis we obtain the
line element
gab dxa dxb = −N 2 dt2 + hab (dxa + N a dt) (dxb + N b dt) (2.8)

in coordinates xa such that ta ∇a = ∂/∂t. This shows that, the space-time

geometry is described not by a single metric but by the spatial geometry of
slices, encoded in hab , together with deformations of neighboring slices with
respect to each other as described by N and N a .
Given a time-evolution vector field, we complete the interpretation of
tensor fields on a foliated space-time as time-dependent tensor fields on
space. Given two spatial slices in the foliation, in order to speak about
time-dependence of the tensor fields, we need to show how tensor fields on
these slices change. To do that, we are first required to uniquely associate
a point on one slice with a point on the other slice, then, by evaluating the
fields at the associated points we can show their changes when going from
one slice to the next. A time derivative of a tensor field is defined as the Lie
derivative along the time-evolution vector field ta :
( )
Ṫ a1 ···anb1 ···bm := hac11 · · · hacnn hbd11 · · · hbdm
m
Lt T c1 ···cn d1 ···dm (2.9)
Canonical General Relativity 4

In the case of a 4-dimensional space-time Einstein’s equation is really

10 different equations, since there are 10 independent components in the
Einstein tensor. We will rewrite these equations in terms of the metric on
the slice Σ, or 3-metric hab , and the ‘extrinsic curvature’ Kab of the slice Σ,
which describes the curvature of the way it sits in M . In what follows we
shall see that the extrinsic curvature can also be thought of as representing
the time derivative of the 3-metric. We can think of (hab , Kab ) as Cauchy
data for the metric, just as we think of the vector potential on space and the
electric field as Cauchy data for electromagnetism or the Yang-Mills field.
We will see that of Einstein’s 10 equations, 4 are constraint equations that
the Cauchy data must satisfy, while 6 are evolutionary equations saying how
the 3-metric changes with time. This is called the Arnowitt-Deser-Misner,
or ADM, formulation of Einstein’s equation.

2.1 Intrinsic and extrinsic geometry

The spatial metric hab itself is an intrinsic quantity, and as a metric it
allows one to define a unique covariant derivative operator Da on Σ such
that Da hbc = 0. This covariant derivative can be written in terms of the
space-time covariant derivative ∇a as

Dc T a1 ···akb1 ···bl := (ha1d1 · · · hakdk hb1e1 · · · hbl el )hc f ∇f T c1 ···cn d1 ···dm . (2.10)

Definition. Given the three dimensional covariant derivative Da , we can

define the intrinsic-curvature tensors as with any covariant derivative:
(3)
Rabcd ωd = Da Db ωc − Db Da ωc (2.11)

for all spatial 1-form ωc , i.e., ωa na = 0. From this intrinsic Riemannian

curvature, we can obtain the intrinsic Ricci tensor (3) Rab and scalar (3) R by
the usual contractions.
.

In contrast to the intrinsic geometry, which applies to a single (Σ, hab ) no

matter how it is embedded in a space-time manifold, the extrinsic geometry
of Σ in R×Σ refers to the bending of Σ in its neighborhood, which in general
implies a changing normal vector field na along Σ. This notion is captured
in the definition of the extrinsic-curvature tensor:
Canonical General Relativity 5

Definition. Given any normal vector na to the surface Σ, the extrinsic-

curvature tensor is a spatial tensor on Σ by definition of Da :

Kab := Da nb = hca hdb ∇c nd . (2.12)

Thus, Kab measures how much the surface Σ is curved in the way it sits in
M , because it says how much a vector tangent to Σ will fail to be tangent
if we parallel translate it a bit using the Levi-Civita connection ∇ on M .
.

The extrinsic-curvature tensor has some properties as follows:

(i) From the definition (2.12) we can write

Kab = hbd hac ∇c nd = (gb d + nb nd )hac ∇c nd

= hac ∇c nb + nb nd hac ∇c nd . (2.13)

It can be seen that, in contrast to commuting hab with Dc , the space-

time metric and its covariant derivative is always commutative.

(ii) The extrinsic curvature tensor is symmetric:

Kab = Kba (2.14)

This also concludes that all spatial projections of ∇a nb are symmetric:

∇a nb − ∇b na = 0.

(iii) From the symmetric property of Kab we have that Kab = 12 (Kab +Kba );
using this together with (2.13) we can write

2Kab = (gac + na nc )∇c nb + (gb c + nb nc )∇c na

= nc ∇c (na nb ) + ∇a nb + ∇b na
= nc ∇c hab + hcb ∇a nc + hac ∇b nc =: Ln hab . (2.15)

Thus, the extrinsic curvature is half of the Lie derivative of the intrinsic
metric along the unit normal:

1
Kab = Ln hab (2.16)
2

(iv) From last identity in Eq. (2.15) we write that

1 c
Kab = [n ∇c hab + hcb ∇a nc + hac ∇b nc ]
2
1
= [N nc ∇c hab + hcb ∇a (N nc ) + hac ∇b (N nc )]
2N
1 c d 1 c d
= ha hb Lt−N hcd = h h (Lt hcd − LN hcd ), (2.17)
2N 2N a b
Canonical General Relativity 6

where we substituted N na = ta − N a and smuggled in projections

hac hbd since Kab is spatial. Furthermore, LN hcd = Da Nb + Db Na .
By using the definition (2.9) we have that ḣab = hac hbd Lt hcd , thus,
Eq. (2.17) can be rewritten as

1 ( )
Kab = ḣab − Da Nb − Db Na (2.18)
2N

Similar to splitting of the space-time metric (2.8) to spatial and temporal

sections, intrinsic and extrinsic curvatures (2.11) and (2.12) can also describe
together the space-time curvature.
One can show that the symmetries of the Riemann tensor Rabcd reduces
number of independent components from n4 , where n is the dimension of
space-time, down to n2 (n2 − 1)/2. For a 4 dimensional space-time, the num-
ber of space-time tensors is 20 and there are 6 components for the spatial
Riemann tensor. As for the Ricci tensor Rab , since it is symmetric one would
expect that it has n(n + 1)/2 independent components. In three dimension
the Ricci tensor has 6 independent components, which is just as many as the
Riemann tensor. In four dimension space-time it has 10 independent compo-
nents. Consequently, using symmetry of the extrinsic curvature, it provides
only 6 components more than the spatial Riemann tensor, which is 12 in-
dependent components. These components we introduced so far constitutes
all curvature components necessary for a canonical decomposition.

2.2 The Gauss-Codazzi equations

We shall show that four of Einstein’s equations are constraints that the
the 3-metric hab and extrinsic curvature Kab must satisfy (see next section).
This is because that some components of the Riemann tensor depend only on
the extrinsic and the intrinsic curvatures, that is, the curvature of hab . The
formulas that describe the precise relations between curvature components
are known as the Gauss-Codazzi equations, which we now derive.
We will compute the components Ref gh in terms of Kab and (3) Rabcd . To
do this we compute
Da Db ωc = Da (hbd hce ∇d ωe ) = haf hbg hch ∇f (hgd hhe ∇d ωe )
= haf hbd hce ∇f ∇d ωe + hce (haf hbg ∇f hgd )∇d ωe
+ hbd (haf hch ∇f hhe )∇d ωe . (2.19)
In the second term we have
haf hbg ∇f hgd = haf hbg ∇f (ggd + ng nd ) = nd hbg ∇a ng = Kab nd ,
and in the last term
hbd (haf hch ∇f hhe )∇d ωe = hbd Kac ne ∇d ωe = −Kac hbd ωe ∇d ne = −Kac Kbe ωe
Canonical General Relativity 7

where we have used ωa na = 0 for spatial ωe . Thus, using definition (2.11)

and the Eq. (2.19) we obtain
(3)
Rabce ωe = Da Db ωc − Db Da ωc
= haf hbd hce (∇f ∇d ωe − ∇d ∇f ωe )
− Kac Kbe ωe + Kbc Kae ωe . (2.20)

This gives the so-called Gauss equation:

hae hbf hcg Ref gh = (3)

Rabcd + Kac Kb d − Kbc Kad (2.21)

By computing the relation

hae hbf hcg Rabcd nd = hae hbf hcg (∇a ∇b − ∇b ∇a )nc

( )
= hae hbf hcg ∇a (gb d ∇d nc ) − ∇b (gad ∇d nc )
= De Kf g − hae hbf hcg ∇a (nb nd ∇d nc )
− Df Keg + hae hbf hcg ∇b (na nd ∇d nc )
= De Kf g − Df Keg − hae hbf hcg (nd ∇d nc )(∇a nb − ∇b na ).

Using the symmetry of the spatial projection of ∇a nb , the last term in this
equation vanishes. The result is the Codazzi equation:

hae hbf hcg Rabcd nd = De Kf g − Df Keg (2.22)

Let us introduce the Ricci equation as

Racbd nc nd = nc (∇a ∇c − ∇c ∇a )nb . (2.23)

This equation can be derived in terms of Ln , the Lie derivative along the
unit normal na , of the extrinsic curvature Kab , and the normal acceleration
aa := nc ∇c na (satisfying aa na = 0):

Racbd nc nd = −Ln Kab + Kac Kbc + D(a ab) + aa ab (2.24)

Using the Ricci equation (2.24), using the relation

Rab na nb = Racdc na nb

we obtain the following equation

Rab na nb = (Kaa )2 − Kab Kb a + ∇a v a (2.25)

where the vector field v a is defined as v a := −na ∇c nc + nc ∇c na .

Canonical General Relativity 8

From the Gauss-Codazzi equations together with the Ricci equation, the
Ricci scalar R reads

R = g ab g cd Rabcd = (hab − na nb )(hcd − nc nd )Rabcd

= hab hcd Rabcd − 2Rab na nb . (2.26)

Then, using the symmetry of the Riemann tensor, we find the Ricci scalar
in terms of the extrinsic curvature

R = (3)
R + Kab K ab − (K aa )2 − 2∇a v a (2.27)

Up to a divergence ∇a v a , we can thus decompose the Ricci scalar into a

“kinetic” term quadratic in extrinsic curvature, and a potential term (3) R
which depends only on the spatial metric and its spatial derivatives.
The extrinsic curvature, as shown by (2.18), plays the role of a “velocity”
of the spatial metric and is thus a candidate for its momentum. In the next
section, we discuss the Hamiltonian formalism of general relativity in terms
of canonical variables.

3 The ADM formalism

The action of general relativity in metric variables is given by Einstein-
Hilbert action
∫ √ ∫
1
SEH [g] = d4 x − det gR =: dt Lgrav (3.1)
16πG
where det g is the determinant of the metric gab . Once the space-time is
foliated, using (2.27) the gravitational Lagrangian in terms of the extrinsic
curvature Kab and the 3-curvature (3) R becomes
∫ ( )
1 √
Lgrav = d3 xN − det h (3) R + Kab K ab − (Kaa )2 (3.2)
16πG

up to boundary terms which do not affect local field equations. The deter-
minant det g = −N 2 det h.

3.1 Constraints
Eq. (3.2) shows that the ten independent components of the space-time
metric gab are replaced by the six components of the induced Riemannian
metric hab on the slice Σ, plus the three components of the shift vector Na
and the lapse function N . The action of general relativity depends on ḣab via
Canonical General Relativity 9

extrinsic curvature Kab , thus, one can obtain the momentum pab conjugate
to hab :
δLgrav 1 δLgrav
pab (x) = =
δ ḣab (x) 2N δKab
√ ( )
det h
= K ab − Kcc hab . (3.3)
16πG
However, the action does not depend on time derivatives of the remaining
space-time metric component N and N a ; therefore, momenta conjugate to
N and N a , are given, respectively, by
δLgrav δLgrav
pN (x) = =0 and pa (x) = =0, (3.4)
δ Ṅ (x) δ Ṅ a (x)
presenting two constraints on the gravitational phase space. Since the rela-
tion (3.3) can be inverted for
16πGN
ḣab = √ (2pab − pcc hab ) + 2D(a Nb) , (3.5)
det h
thus, relations in (3.4) present two primary constraints. Then one can work
out the total Hamiltonian by the formula
∫ ( )
Hgrav = d3 x ḣab pab + λpN + µa pa − Lgrav , (3.6)

which, by substituting ḣab from (3.5), leads to

∫ [ ( )
16πGN 1 a 2
Hgrav = d x √
3
pab p − (pa ) + 2pab Da Nb
ab
det h 2
√ ]
N det h (3)
− a
R + λpN + µ pa . (3.7)
16πG

The primary constraints (3.4) imply secondary constraints:

0 = ṗN = {pN , Hgrav } =: −Cgrav (hab , pab ) , (3.8)

0 = ṗN a = {pN a , Hgrav } =: −Cagrav (hab , pab ) . (3.9)

By working out the Poisson bracket in these relations we obtain the so-called
Hamiltonian constraint:
( ) √
16πG 1 a 2 det h (3)
Cgrav = √ pab p − (pa ) −
ab
R ≈ 0 (3.10)
det h 2 16πG

and the diffeomorphism constraint:

Cagrav = −2Db pba ≈ 0 (3.11)

Canonical General Relativity 10

∫ √ √
We have integrated by parts, using 2 d3 x det hDa (pab Nb / det h) as a
boundary term for any vector field N a , in derivation of Cgrav .
The lapse function N and the shift vector N a now play the role of La-
grange multipliers of the secondary constraints. Then, with these notations,
the total Hamiltonian, as a linear combination of the constraints, can be
written as where we identify the Hamiltonian density as follows
∫
Hgrav = d3 x (N Cgrav + N a Cagrav + λpN + µa pN a ) + H∂Σ (3.12)

where H∂Σ is the Hamiltonian of the boundary term.

The fact that the Hamiltonian involves terms proportional to the lapse
and shift should not be surprising, since the role of the Hamiltonian is to
generate time evolution, and in general relativity we need to specify the lapse
and shift to know the meaning of time evolution. However, if we express
the quantities Cgrav and Cagrav in terms of the extrinsic curvature using the
formula following Gauss-Codazzi equations, we find that
Cgrav = −2Gab na nb and Cagrav = −2Gai na
This implies that the Hamiltonian density for general relativity must vanish
by the vacuum Einstein equation: Hgrav = 0. In other words, the constraints
Cgrav = Cagrav ≈ 0 are precisely the 4 Einstein equations that are constraints
on the initial data. Therefore, there is no proper Hamiltonian which would
be non-trivial on the constraint surface. This is in agreement with the fact
that there is no absolute time in general relativity, since a non-vanishing
Hamiltonian would generate time evolution in an external time parameter.
Instead, dynamics is determined by the constraints, such that evolution
as a gauge flow can be parametrized arbitrarily. In this way, we see the
reparameterization invariance of coordinates in a generally covariant thoery.
Since in the herein canonical formalism, the configuration space of gen-
eral relativity is M et(Σ), thus, it is natural to expect that the phase space
Γ is the space of all pairs (hab , N, N a ; pab , pN , pN a ), or the cotangent bundle
T ∗ M et(Σ). However, not all points of this phase space represent allowed
states. The Einstein equations that are constraints must be satisfied, and
this restriction picks out a subspace of the phase space called the physical
phase space:
Γphys = {Cgrav ≈ 0; Cagrav ≈ 0} ⊂ T ∗ M et(Σ) . (3.13)
The Hamiltonian (3.12) vanishes on this subspace. In section 3.3 we will see
that these constraints are first class and thus generate gauge transformations
which do not change the physical information in solutions. The Hamiltonian
constraint does this for time, and the diffeomorphism constraint for spatial
coordinates. Once these constraints are satisfied, we make sure that the
formulation is space-time covariant even though we started the canonical
formulation with slicing of space-time determined by any time function t.
Canonical General Relativity 11

3.2 Equations of motion

In the presence of the total Hamiltonian of general relativity, we can obtain
the evolutionary part of Einstein’s equations by this means. These are really
just the equations Gab = 0 in disguise, which are equations for the second
time derivative of the 3-metric, but rewritten so as to give twice as many
first-order equations. Then, Hamiltonian equations give Ṅ (x) = λ(x) and
Ṅ a (x) = µa (x), which tells us that these functions can change arbitrarily
due to reparameterizations. Moreover,

ḣab = {hab , Hgrav } and ṗab = {pab , Hgrav } .

The first relation just reproduces the equation (3.5) in terms of the momen-
tum. Finally the last relation is a non-trivial evolution equation which can
be computed in several steps. We write only the final equation here (for
details of calculation see [2]):
δHgrav
ṗab = {pab , Hgrav } = −
δhab
√ ( ) ( )
N det h (3) ab 1 (3) ab 8πGN ab cd 1 c 2
= − R − Rh + √ h p pcd − (pc )
16πG 2 det h 2
( ) √ ( )
32πGN 1 det h
− √ pac pcb − pab pcc + Da Db N − hab Dc Dc N
det h 2 16πG
√ ( √ )
+ det hDc pab N c / det h − 2pc(a Dc N b) . (3.14)

The point is that, even on the physical phase space where Hgrav = 0, the
time evolution given by Hamilton’s equations is nontrivial.
If matter sources are present, they, too, contribute to the action and thus
to the canonical constraints. In particular, the matter Hamiltonian Cmatt
will be added to the Hamiltonian constraint Cgrav , and energy flows of matter
will be added to the diffeomorphism constraint Cagrav . Then, the combined
Hamiltonian constraint is C = Cgrav + Cmatt and the total diffeomorphism
constraint reads Ca = Cagrav +Camatt . It is convenient to exhibit the structure
of the system for constraints in smeared form, integrated with respect to
the multipliers N and N a : the smeared Hamiltonian and diffeomorphism
constraints are given, respectively, by
∫ ∫
3
H[N ] := d xN (x)C(x) = d3 xN (Cgrav + Cmatt ) , (3.15)
∫ ∫
( )
D[N a ] := d3 xN a (x)Ca (x) = d3 xN a Cagrav + Camatt . (3.16)

Recall that the lapse and shift measure how much time evolution pushes
the slice Σ in the normal direction and the tangent direction, respectively.
In particular, if we set the shift equal to zero, the Hamiltonian for general
Canonical General Relativity 12

relativity is equal to H[N ], and it generates time evolution in a manner that

corresponds to pushing Σ forwards in the normal direction. On the other
hand, if we set the lapse equal to zero, the Hamiltonian becomes D[N a ],
which generates a funny sort of ‘time evolution’ that pushes Σ in a direction
tangent to itself. More precisely, this quantity generates transformations of
(total) physical phase space Γ = Γgrav × Γmatt corresponding to the flow on
Σ generated by N a . This flow is a 1-parameter family of diffeomorphisms
of Σ. For this reason, Ca or D[N a ] is called the diffeomorphism constraint,
while C or H[N ] is called the Hamiltonian constraint. It is actually no
coincidence that C and Ca , play a dual role as both constraints and terms
in the Hamiltonian. This is, in fact, a crucial special feature of field theories
with no fixed background structures.

3.3 General constraint algebra

So far, we have seen that in general relativity, there are four primary con-
straints pN ≈ 0 and pN a ≈ 0, whose time derivatives lead to more four
secondary constraints C = Cgrav + Cmatt ≈ 0 and Ca = Cagrav + Camatt ≈ 0.
Since the eight constraints pµ := (pN , pN a ) and Cµ := (C, Ca ) are inde-
pendent, thus, they form eight first-class constraints for general relativity.
Thus, these constraints form a first-class algebra in which there are four
independent gauge transformations by changing space-time coordinates, ex-
actly the number of secondary constraints on phase-space functions. (The
primary constraints only generate change of N and N a .) The total Hamilto-
nian (3.12) is a linear combination of (first class) constraints, i.e., it vanishes
identically on solutions of equations of motion. This is a generic property
of generally covariant systems.
In addition to the first-class nature, which tells us that Poisson brack-
ets of the constraints vanish on the constraint surface, there is a specific
“off-shell algebra” of constraints, satisfied by the constraint functions on
the whole phase space including the part off the constraint surface: this
shows what kinds of transformation the constraints generate, and how they
are related to space-time properties. In the presence of the matter contri-
bution it is interesting to see the Poisson brackets of the Hamiltonian and
diffeomorphism constraints. The full constraint algebra is then given as

{D[N b ], D[M a ]} = D[LN b M a ] , (3.17)

{D[N ], H[N ]} = H[LN a N ] ,
a
(3.18)
{H[N ], H[M ]} = −D[hab (N ∂b M − M ∂b N )]. (3.19)

The relations above are know as Dirac algebra. In the first two lines, we sim-
ply have the expected action of infinitesimal spatial diffeomorphisms, with
multipliers on the right-hand side given by Lie derivatives LN b M a = [N, M ]a
and LN a N = N a ∂a N . The last line, includes the phase space function hab ,
Canonical General Relativity 13

which is the so-called structure function, rather than just phase-space inde-
pendent structure constants.

References
[1] John C. Baez, Javier P. Muniain, Gauge Fields, Knots and Gravity,
(Series on Knots and Everything, Vol 4, 1994).

[2] Martin Bojowald, Canonical gravity and applications (Cambridge Uni-

versity Press, 2011).