VOLUME 57, NUMSER 18 PHYSICAL REVIEW LETTERS 3 NovEMsER 1986

New Variables for Classical and Quantum Gravity

Abhay Ashtekar
Physics DepartmentS, yracuse University, Syracuse, IVew York 13244, and Institute for Theoretical Physics,
University of California, Santa Barbara, Santa Barbara, California 93106
(Received 18 Ibm:mber 1985; revised manuscript received 29 August 1986)

A Hamiltonian formulation of general relativity based on certain spinorial variables is introduced.

These variables simplify the constraints of general relativity considerably and enable one to imbed the
constraint surface in the phase space of Einstein's theory into that of Yang-Mills theory. The imbedding
suggests ne~ ways of attacking a number of problems in both classical and quantum gravity. Some illus-
trative applications are discussed.

PACS numbers: 04.60.+n, 04.20.Fy

Attempts at constructing perturbative quantum gravity tangent space of Z and makes them SU(2) spinor indices.
have been unsuccessful. It is now generally believed that Furthermore, each soldering form, ct, is a "square root"
the problem lies in the basic assumption of the perturba- of a (positive definite) metric q, b on Z,
tion theory that the true space-time structure can be well AB MN
gab. a &b gM BN Tr &a &b,
approximated by a classical background geometry even
below the Planck scale. From this standpoint, there is lit- and singles out a unique torsion-free connection D on ten-
tle hope in retaining the general perturbative framework sor and spinor fields on Z satisfying
and simply changing, e.g. , the form of the Einstein
Lagrangean by adding higher-derivative terms or super- D ega 0, D ct tt 0. (2)
symmetric matter. A more promising direction is to face The configuration space 8 for general relativity is to
the problem nonperturbatively. For, as has been em- be the space of all (suitably) regular and, if Z is noncom-
phasized by 3. Klauder over the years, quantum gravity
pact, asymptotically well-behaved soldering forms a'"tt.
may well exist as an exact theory in spite of perturba- The phase space I is the cotangent bundle over 8. Thus,
tive nonrenormalizability. The canonical quantization a point of I consists of a pair (a'"tt, M~ tv), where M, a
scheme provides a natural avenue in this direction since it density of weight 1, is the momentum canonically conju-
does not require the fixation of a classical background
gate to o. The canonical variables (q,b, p' ) of the tradi-
geometry. Furthermore, the fact that the Hamiltonian tional Hamiltonian formulation' are now to be regarded
structure of general relativity has certain essentially non- as "derived" quantities: q, b is given by (1) and p' by
perturbative features — in the exact theory, the Hamil-
tonian is essentially given by the constraints, while the +ab. ~ Tr~ &(a~b)m —~(ab) (3)
two decouple in any order in perturbation theory —
sug-
gests that qualitatively new results may arise from exact As usual, not all points of I are accessible to the physical
canonical quantization. gravitational field: There are constraints. First, we have
Over the years, the main obstacle to the canonical the familiar constraints
quantization program has come from the fact that the
constraint equations have a complicated, nonpolynomial Cb(a M): -D.p"-0 (4)
dependence on the traditional canonically conjugate vari-
ables. The purpose of this Letter is to report the ex-
istence of new variables in terms of which the constraints
c(~,w): -(p'p. b —, p')—— (de'tq )
(5)
simplify considerably and to point out that the use of
these variables provides new, nonperturbative approaches ~here R is the scalar curvature of q, b. However, since
to problems in both classical and quantum gravity. we have enlarged the configuration space from the space
Let us begin with some mathematical preliminaries. of the six-component fields q, b to that of the nine-
Fix a three-manifold X and consider on it, in addition to component fields o "tt, we have three new constraints:
tensor fields, T'", s, fields with "internal" SU(2) in- m'b'-0.
dices, X,",It~, . . . . The SU(2) structure provides us with
volume forms, s" and a~a, on internal indices, satisfying The canonical transformations generated by these con-
c"ae~D b D. We shall use them to raise and lower straints cause SU(2) rotations on the internal indices of
these indices: A,":-a" ktt and It~.. p ett~. Now, given o and M.
any isomorphism erg" between tangent vectors V' and The above framework is equivalent to the familar triad
second-rank, trace-free, Hermitian fields V"tt, the identi- formalisms. The key new step is the introduction of new
fication V'=——ag"Vaz solders the internal indices to the variables on I.
Given any point (ct, M) of introduce I,
1986 The American Physical Society
VOLUME 57, NUMBER 18 PHYSICAL REVIEW LETTERS 3 NOVEMBER 1986

two connections —S on Z: cidentally, automatically preserve the Ansatz). The re-

sulting system of equations provides a new, simple, and
a aha. Da abM —(i l ~»&est"abrv convenient characterization of self-dual solutions. Tra-
where H, ~ is given by2 ditionally, 0-space and twistor techniques have been used
to study these solutions. The new variables serve to
bridge these techniques to the Hamiltonian methods.
=G(detq) '"(M.st~ .
,' o—st—"a'„ahab"a) Other applications, to classical relativity, include the fol-
lowing: analysis of gravitational perturbations; interest-
The use of these connections simplifies the structure of ing, exact solutions to constraint equations; understand-
constraints (4) and (5) considerably. To see this, intro- ing and generalizing of the results5 on the relation be-
duce connection one-forms —A, and curvature two-forms tween certain classes of solutions to Einstein's and
F,b ass—ociated with — S: Yang-Mills equations; and the use and role of hypersur-
face twistors in general relativity.
—S,as' .8, a~+G —A, sr aiv, On the phase space I", the new variables have several
+ + + interesting properties. Each of [+A,], [ A, ], and
fa +blast:G Faber aw
= (detq)'~~a ] forms a complete set of commuting
[cr'—
where rl is a fixed (c number) connection, also satisfying variables with respect to the natural Poisson bracket on
8 e~a 0. Then (6) can be rewritten as I . Furthermore, —A, is "conjugate" to a in the sense
that
(6')
[-A ~"(x),&ga(y)]p. a.
and, modulo (6), Eqs. (4) and (5) can be recast as
Tres Fs 0, — (4') [The factor of G in (9) ensures that (A) (8) has dimen-
Tro a 0. — (5') sions of action. ] shall therefore use 8' and —A, as the
I
F,i,
basic variables. [Note that one can replace a by 8' in
(Throughout, ~ stands for + or —;we can use either the constraints (4')-(6') free of charge. ] These proper-
+A, or A, .) Constraints (4')-(6') are closed under the ties suggest identifications with certain variables that are
Poisson bracket and preserved by dynamics. featured in the Yang-Mills theory. —A, is the connec-
Note that the form of constraints (4')-(6') is simpler tion one-form; — 8': e'+ Fs„ its —magnetic field; and
than that of (4)-(6) in at least two respects. First, a', the analog of the electric field E'. In terms of these
(4')-(6') are at most quadratic in each of the new vari- Yang-Mills variables, the constraints become
ables (a, —A, ) while (4)- (6) involve nonpolynomial
—S E 0, (6")
functions of q, b. Second, if one were to regard —A, as
the new configuration variable and a as the "momen- (4")
tum,
" (5') involves only a "kinetic" term, quadratic in
TrEx B 0,
the new momenta, and is therefore structurally similar to TrE (ExB) 0. (5")
the strong-coupling limit of (5) in which the "potential"
Thus, every initial datum (satisfying constraints) (a, M)
term, R, is neglected. [This came about because —A,
for Einstein's equation yields initial data (A, E) for
knows both about p' and (the connection of) q, b. ]
Yang-Mills equations which, in addition, satisfy four
These features lead to new avenues especially in the
constraints which are purely algebraic in field strength;
quantum theory.
one has an imbedding of the Einstein constraint surface
What is the physical interpretation of —A, '? Consider
into the Yang-Mills theory. This imbedding preserves
a solution of g, b of Einstein's equation obtained from ini-
the Poisson-bracket structure of the two theories. On the
tial data (a', M, ) [satisfying (4)- (6)]. Then one has the
other hand, it does not commute with time evolution; the
following: — are the restrictions to X of the four-
S Yang-Mills Hamiltonian is very different from Einstein s.
dimensional spin-connection V on (un)primed SL(2, C)
spinors (e.g. , +$,
1st q, VP, ), and Tr F,sa'e'
'
"— However, since the Einstein Hamiltonian is a linear com-
bination of constraints and a surface term, the simplifica-
-(J2/G)(E' -ti8'"), where F. and are the elec-8' tion of constraints is significant also for Einstein dynam-
tric and magnetic parts, relative to X, of the %eyl tensor

:
1cs.
of g, b. Thus, —A, is a potential for the (anti-)self-dual To go over to quantum theory, as in the Yang-Mills
part of the %eyl tensor. This fact leads to an interesting case, we shall replace the basic variables —A, and
application: One can obtain (complex Lorentzian or, 8'— E' by operators —A, and o satisfying the canonical
with minor convention changes, real Euclidean) self-dual commutation relation
solutions to Einstein's equation by simply setting
+A, 0. This Ansatz trivializes (4') and (5') and simpli- [ —A ~n(x), a"ga(y)]
fies (6') as well as the evolution equations (which, in- -(P, t JZ)b a„'"b,"'b(x,y). (»)
VOLUME 57, NUM@BR 18 PHYSICAL REVIEW LETTERS 3 NovEMaER 1986

The shift of variables from (q, p) to (cr, A—) simphfies several issues in the quantum theory as a result of the features of
constraints (4')-(6'), noted below (5'). I now summarize the construction and the results that follow.
First, one can ask if there exists a factor ordering for the quantum version of constraints (4')-(6') for which the quan-
tum constraints are closed under the commutator bracket, i.e., for which the evaluation of commutators yields a result in
which a constraint operator always appears on the right. The answer is in the affirmative. Set

Ctv. '(J2/h)„G Ng ( —S~cr")tt +N'Tra —F,b+NTrcr'cr F,b-,

where N stands for the triplet of smearing fields (N'tt, N', N ). By use of (12) it then follows that

[&tv, cbr] -Ct,

where

P~a [N, M]ga+G 'N~Mb F,bg —+G '(MN —NM')(cr g F,bbt —

—cr bi F,bg ),
(i4)
P' Xbt¹+2(NDbM —MDbN )q', P XbtN —LrtM.
In this result, the presence of internal indices and the
t
consequent constraint (6') plays a crucial role: Certain simple harmonic oscillator). In exact theory the repre-
unwanted terms vanish because of the symmetry proper- sentation is being investigated by Jacobson and Smolin.
ties of their internal indices, and the commutators of (4') Next, note that the constraints (4")-(6") in terms of
and (S') involve not only these constraints but also (6'). the Yang-Mills variables ( A„E')— do not require a
Also, the presence of the internal indices in (4')-(6') (as background structure such as a metric or a derivative
well as of cr', the "square root" of q ) makes it impossi- operator. (This is to be contrasted with the Yang-Mills
ble to translate the preferred factor ordering in terms of evolution equations which do require a background
the traditional variables (q, b,p' ). Note, however, that metric. ) Consequently, one can take over techniques
the argument is oniy formal; I have not regularized the from the Hamiltonian lattice QCDs to put the quantum
products of operator-valued distributions in (4')-(6'). gravity on a lattice. The advantage of a lattice formula-
Nonetheless, formal closure is significant since it can fail tion is that the constraints do not have to be regularized.
even for systems with a finite number of degrees of free- This line of investigation is being pursued by Renteln and
doms where the issues of regularization never arise. Smolin.
Next, we come to the issue of finding representations Finally, I have restricted the discussion to the vacuum
of the canonical commutation relation (12). Since, as case only for simplicity. It is straightforward to include a
noted above, the constraints are at worst quadratic in cosmological constant and matter sources —Yang-Mills
each of cr and A„ it is feasible to study both the cr repre- —
sources fit in especially well in the framework.
sentation, in which quantum states are general complex- Details will appear elsewhere.
valued functionals of cr, and the —A, representation, in I am most grateful to Ted Jacobson, Lee Smolin, and
which they are holomorphic functionals of —A, . [The Paul Renteln for discussions. This work was supported
analogs for a simple harmonic oscillator are respectively by National Science Foundation Grants No. PHY-83-
the position representation, ter-= ter(x), and the Bargmann 10041 and No. PHY82-17853, supplemented by funds
representation, 1Jr=~(z), z x+ip ]This i.s in striking from the U. S. National Aeronautics and Space Adminis-
contrast to the situation with (q, b,p'b) variables, where tration.
the momentum representation is unmanageable because
the constraints have a complicated q dependence.
Let us focus on the —A, representation since it enables
one to borrow some ideas from (quantum) Yang-Mills
theory. The weak-field hmit has been studied in detaiL
Here, the quantum constraints are solved precisely by the tSee, e.g. , K. Kuchar, in Quantum Gravity 2, edited by C. J.
requirement that the states be holomorphic functionals of Isham, R. Penrose, and D. W. Sciama (Oxford Univ. Press,
the symmetric, trace-free, transverse part (bA,')~~ of New York, 1981).
the linearized connection (b —A, ), and the Hamiltonian
is given simply by
zNote that, when (6) holds,
ture.
~ is just the extrinsic curva-

3A. Ashtekar, T. Jacobson, and L. Smolin, to be published.

4See, e.g. , M. Ko, M. Ludvigsen, E. T. Newman, and K. P.
H-(G/4~)J ( ~h) b'( ~b)$&& d'x Tod, Phys. Rep. 71, 51 (1981).
sR. S. Ward, "Integrable and Solvable Systems" (to be pub-
lished).
(which is analogous to the expression 0 ZZ for the ~See, e.g. , A. Ashtekar and M. Stillerman, J. Math. Phys.
VOLUME 57, NUMBER 18 PHYSICAL REVIEW LETTERS 3 NaVEMSER 1986

(N. Y.) 27, 1319 (1986). states with a large number of loops. (T. Jacobson and L. Smo-
7They construct quantum states from the holonomy operator lin, private communication. )
of —A associated with closed loops and regularize the action of See, e.g. , L. Susskind, in 8'eak and E/ectromagnetic In-
constraints on these loop states. Their results indicate that the teractions, edited by R. Balian and C. H. Llewellyn Smith
metric q' would be degenerate microscopical, ly and acquire its (North-Holland, Amsterdam, 1977).
usual geometrical meaning only on macroscopic regions and in 9P. Renteln and L. Smolin, private communication.