1: \section{The Renormalization Group Equation}
2: \label{sect:RGE}
3:
4: We are finally in a position to write down the renormalization group
5: equation explicitly and study some of its properties.
6: In the $\alpha$--representation, this is given by eq.~(\ref{RGEA0})
7: with the coefficients $\eta$ and $\nu$ from eqs.~(\ref{ETA}) and
8: (\ref{NU}), respectively.
9: We summarize here these equations for convenience:
10: \be\labe{RGEA}
11: {\del W_\tau[\alpha] \over {\del \tau}}\,=\,
12: \left\{ {1 \over 2} {\delta^2 \over {\delta
13: \alpha_\tau^a(x_\perp) \delta \alpha_\tau^b(y_\perp)}}
14: [W_\tau\eta_{xy}^{ab}] -
15: {\delta \over {\delta \alpha_\tau^a(x_\perp)}}
16: [W_\tau\nu_{x}^a] \right\}\,,
17: \ee
18: where
19: \be\label{eta}
20: \eta^{ab}(x_\perp,y_\perp)
21: &=&{1\over \pi}\int {d^2z_\perp\over (2\pi)^2}\,
22: \frac{(x^i-z^i)(y^i-z^i)}{(x_\perp-z_\perp)^2(y_\perp-z_\perp)^2 }
23: \Bigl\{1+ V^\dagger_x V_y-V^\dagger_x V_z - V^\dagger_z V_y\Bigr\}^{ab},
24: \\
25: \nu^a (x_\perp)&=&{ig\over 2\pi}
26: \int {d^2z_\perp\over (2\pi)^2}\,\frac{1}{(x_\perp-z_\perp)^2}
27: \,{\rm Tr}\Bigl(T^a V^\dagger_x V_z\Bigr).
28: \label{nu}\ee
29: Note that, as compared with the original equations (\ref{RGEA0}),
30: (\ref{ETA}) and (\ref{NU}), we have rescalled here the coefficients
31: $\eta$ and $\nu$ to absorb the overall factor $\alpha_s=g^2/4\pi$.
32: The equations above represent our main result in this and the
33: accompanying paper \cite{PI}.
34: They govern the flow with $\tau=\ln(1/x)$ of the probability density
35: $W_\tau[\alpha]$ for the stochastic color field $\alpha_a(\vec x)$
36: which describes the CGC in the COV-gauge. We now turn to a systematic
37: discussion of the properties of these equations.
38:
39: \subsection{General properties}
40: \label{sect:PROP}
41:
42: {\sf i) The coefficients $\eta$ and $\nu$ are real quantities.}
43: This is so since the Wilson lines in the
44: adjoint representation are real color matrices:
45: $V^*_{ab}=V_{ab}$, and therefore $V^\dagger_{ab} = V_{ba}$.
46: By using this same property, one can further verify that:
47:
48: {\sf ii) The 2-point function
49: $\eta^{ab}(x_\perp,y_\perp)$ is symmetric and positive semi-definite: }
50: \be\label{etasym}
51: \eta^{ab}(x_\perp,y_\perp)&=&\eta^{ba}(y_\perp,x_\perp)\nn
52: &=&{1\over \pi}\int {d^2z_\perp\over (2\pi)^2}\,
53: \frac{(x^i-z^i)(y^i-z^i)}{(x_\perp-z_\perp)^2(y_\perp-z_\perp)^2 }
54: \Bigl(1- V^\dagger_z V_x\Bigr)^{ca}
55: \Bigl(1- V^\dagger_z V_y\Bigr)^{cb}.\ee
56: This guarantees that the solution $W_\tau[\alpha]$ is
57: positive semi-definite (as it should) at any $\tau$
58: provided it was like that at the initial ``time'' $\tau_0$.
59:
60: {\sf iii) The RGE preserves the
61: correct normalization of the weight function:}
62: \be\label{norm1}
63: \int {\cal D}\alpha\, \,W_\tau[\alpha]\,=\,1\qquad
64: {\rm at\,\, any}\,\, \tau.\ee
65: Indeed, the r.h.s. of eq.~(\ref{RGEA})
66: is a total derivative with respect to $\alpha$. Thus, if
67: eq.~(\ref{norm1}) is satisfied by the initial condition at
68: time $\tau_0$, then it will be automatically true at any $\tau>\tau_0$.
69:
70: {\sf iv) The solution $W_\tau[\alpha]$ encodes the information about the
71: longitudinal support of $\alpha^a(x^-,x_\perp)$ and its evolution with
72: $\tau$.} From the calculation of quantum corrections, we know that
73: the quantum evolution up to $\tau$ generates a non-trivial color field
74: $\alpha^a(\vec x)$ only within the range $0\le x^-\le x^-_\tau$,
75: with $x^-_\tau=x^-_0{\rm e}^{\tau}$ and $x^-_0= 1/P^+$. Thus, as a functional
76: of $\{\alpha^a(x^-,x_\perp)\,|\,0\le x^-< \infty\}$, $W_\tau[\alpha]$ must
77: have the following structure:
78: \be\label{longitW}
79: W_\tau[\alpha]\,=\,\delta[\alpha_>]\,
80: {\cal W}_\tau[\alpha_<].\ee
81: Here, $\alpha_<$ ($\alpha_>$) is the function $\alpha(\vec x)$
82: for $x^-< x^-_\tau$ ($x^-> x^-_\tau$),
83: \be
84: \alpha(\vec x)\,\equiv\,\theta(x^-_\tau-x^-)\alpha_<(\vec x)\,+\,
85: \theta(x^- - x^-_\tau)\alpha_>(\vec x),\ee
86: the {\it functional} $\delta$--function $\delta[\alpha_>]$
87: should be understood
88: in terms of some discretization of the longitudinal axis,
89: as the product of ordinary $\delta$--functions at all
90: the points $x^->x^-_\tau$ :
91: \be
92: \delta[\alpha_>]\,\equiv\,\prod_{x^->x^-_\tau}
93: \delta(\alpha_{x^-})\ee
94: (this simply shows that $\alpha^a(x^-,x_\perp)=0$
95: with probability one for any $x^->x^-_\tau$),
96: and the new functional ${\cal W}_\tau[\alpha_<]$ involves
97: $\alpha^a(x^-,x_\perp)$ only within the restricted
98: range $0\le x^-\le x^-_\tau$.
99:
100: The structure (\ref{longitW}) can be used to simplify the
101: average over $\alpha$ when computing observables:
102: \be\label{OBSERV}
103: \langle O[\alpha] \,\rangle_\tau&=&
104: \int_{0\le x^-< \infty} \,{\cal D}\alpha\,O[\alpha] \,W_\tau[\alpha]\nn
105: &=&\int_{0\le x^-\le x^-_\tau}\,{\cal D}\alpha\,O[\alpha] \,
106: {\cal W}_\tau[\alpha],\ee
107: where in the first line the functional integral
108: runs over fields with support at $x^- \ge 0$, while in the second
109: line the support is restricted to $x^-\le x^-_\tau$. The
110: expression in the first line is nevertheless more convenient to
111: derive an evolution equation for $\langle O[\alpha] \,\rangle_\tau$,
112: since there the whole dependence on $\tau$ is carried by the weight
113: function:
114: \be\labe{evolO}
115: {\del \over {\del \tau}}\langle O[\alpha] \,\rangle_\tau\,=\,\int
116: {\cal D}\alpha\,O[\alpha] \,{\del W_\tau[\alpha] \over {\del \tau}},\ee
117: with $\del W_\tau/\del \tau$ determined by the RGE (\ref{RGEA}).
118:
119: {\sf v) Within the RGE,
120: %(or, more generally, whenever multiplying $W_\tau[\alpha]$)
121: the Wilson lines $V^\dagger$ and $V$,
122: which were a priori defined as path-ordered
123: integrals along the whole $x^-$ axis
124: (cf. eq.~(\ref{v})), can be effectively replaced by:}
125: \be\labe{vtau}
126: V^\dagger(x_{\perp})\,\rightarrow\,U^\dagger_\tau(x_{\perp}),\qquad
127: U^\dagger_\tau(x_{\perp})\,\equiv\,{\rm P} \exp
128: \left \{
129: ig \int_{0}^{x^-_\tau} dx^-\,\alpha (x^-,x_{\perp})
130: \right \}.\ee
131: This obvious consequence of eq.~(\ref{longitW}) is interesting since
132: it shows that the functional derivatives within the RGE (\ref{RGEA})
133: act on Wilson lines as derivatives with respect to the color field
134: at the {end} point $x^-=x^-_\tau$. Explicitly:
135: \be\label{DIFFU}
136: {\delta U_\tau^\dagger (x_{\perp})\over \delta \alpha^a_\tau(z_\perp)}\,=\,
137: ig\delta^{(2)}(x_{\perp}-z_\perp)\,T^a U_\tau^\dagger (x_{\perp}),\quad
138: {\delta U_\tau(y_{\perp})\over \delta \alpha^a_\tau(z_\perp)}\,=\,
139: -ig \delta^{(2)}(y_{\perp}-z_\perp)\, U_\tau(y_{\perp})T^a\ee
140: which should be compared with more general formulae like:
141: \be\label{DIFFU1}
142: {\delta V^\dagger (x_{\perp})\over \delta \alpha^a_\tau(z_\perp)}\,=\,
143: ig\delta^{(2)}(x_{\perp}-z_\perp)\,U_{\infty,\tau}^\dagger(x_{\perp})
144: T^a U_\tau^\dagger (x_{\perp})\,,\ee
145: where the integral over $x^-$ within $U_{\infty,\tau}^\dagger$
146: runs from $x^-_\tau$ up to $\infty$.
147: Clearly, eq.~(\ref{DIFFU1}) reduces to the first eq.~(\ref{DIFFU})
148: for a field $\alpha$ with support at $x^-\le x^-_\tau$.
149:
150: %{\sf vi) The non-commutativity of the second order mixed derivatives
151: %is harmless.} A straightforward application of the differentiation
152: %rules (\ref{DIFFU}) shows that the second order
153: %mixed derivatives do not commute with each other. We rather have
154: %(with transverse coordinates omitted, for simplicity):
155: %\be\label{DIFFMIX}
156: %{\delta \over \delta \alpha^b_\tau}\,
157: %{\delta U_\tau^\dagger \over \delta \alpha^a_\tau}\,=\,-g^2 T^a T^b
158: %U_\tau^\dagger\,,\qquad {\delta \over \delta \alpha^a_\tau}\,
159: %{\delta U_\tau^\dagger \over \delta \alpha^b_\tau}\,=\,-g^2 T^b T^a
160: %U_\tau^\dagger\,,\ee
161: %to be compared with (in obvious notations):
162: %\be\label{DIFF2}
163: %{\delta^2 V^\dagger
164: % \over {\delta
165: %\alpha^a(x^-) \delta \alpha^b(y^-)}}\,=\,-g^2\Bigl\{
166: %\theta(x^--y^-)U_{\infty,x^-}^\dagger T^a U_{x^-,y^-}^\dagger T^b
167: % U_{y^-}^\dagger \nn
168: %+ \theta(y^--x^-)U_{\infty,y^-}^\dagger T^b U_{y^-,x^-}^\dagger T^a
169: % U_{x^-}^\dagger\Bigr\}\,,\ee
170: %which is symmetric, but discontinuous at $x^- = y^-$. Clearly, it
171: %is this latter discontinuity which is responsible for the lack
172: %of commutativity of the mixed derivatives at the end point: If, e.g., the
173: %argument of the first derivative is already the end point $x^-=x^-_\tau$,
174: %then the second derivative, with argument $y^-$, can act only on the field
175: %at $y^- \le x^-_\tau$. Thus, in this sequence, the limit $x^-=y^-= x^-_\tau$
176: %can be achieved only as $x^-,\,y^- \to x^-_\tau$ with $y^- < x^-$; this
177: %selects indeed the first term in the r.h.s. of eq.~(\ref{DIFF2}), in
178: %agreement with the first equality (\ref{DIFFMIX}).
179: %Note, however, that this lack of commutativity is harmless at the level of
180: %the RGE, since the second order derivative there is multiplied
181: %by the {\it symmetric} kernel $\eta^{ab}$ (cf.
182: %eq.~(\ref{etasym})).
183:
184: {\sf vi) The longitudinal coordinate and the evolution time
185: get identified by the quantum evolution.} The previous arguments
186: at points {iv) -- vii)} reveal a strong correlation
187: between the longitudinal coordinate $x^-$ and the rapidity $\tau$
188: which eventually allows us to identify these two variables.
189: To make this identification more precise, let us
190: introduce, in addition to the
191: {\it momentum} rapidity $\tau\equiv\ln(P^+/\Lambda^+) = \ln(1/x)$,
192: also the {\it space-time} rapidity ${\rm y}\,$,
193: defined as (for positive $x^-$ and $x^-_0= 1/P^+$):
194: \be\label{etarapdef} {\rm y}\,\equiv\,\ln(x^-/ x^-_0)\,,
195: \qquad -\infty < {\rm y} < \infty\,.\ee
196: Then the field
197: \be\label{alphatau}
198: \alpha_{{\rm y}}^a(x_\perp)
199: \,\equiv\, \alpha^a(x^-=x^-_{\rm y},x_\perp),\qquad
200: x^-_{\rm y}\,\equiv\,x^-_0{\rm e}^{{\rm y}},\ee
201: at {\it positive}\footnote{The field $\alpha_{{\rm y}}$ at
202: negative ${\rm y}$, i.e., at $x^-<x^-_0$, is rather a part of the
203: initial conditions, in that it exists independently of the quantum evolution.}
204: ${\rm y}$ (i.e., at $x^->x^-_0$) has been
205: generated by the quantum evolution from $\tau'=0$ up to $\tau$,
206: with a one-to-one correspondence between ${\rm y}$ and $\tau$.
207: That is, the field $\alpha_{{\rm y}'}$ within the {\it space-time} rapidity
208: layer ${\rm y}\le {\rm y}' \le {\rm y}+\Delta{\rm y}$ has been
209: obtained by integrating out the quantum modes within the
210: {\it momentum} rapidity layer $\tau\le \tau'\le \tau+\Delta\tau$
211: with $\tau={\rm y}$ and $\Delta\tau=\Delta{\rm y}$.
212: This leads us to treat ${\rm y}$ and $\tau$ on the same
213: footing, as an ``evolution time''. Then $\{\alpha_{{\rm y}}^a(x_\perp)\,|\,
214: -\infty < {\rm y}<\infty\}$ is conveniently interpreted as a {\it trajectory}
215: in the functional space spanned by the {\it two-dimensional} color fields
216: $\alpha^a(x_\perp)\,$; this trajectory effectively ends up at ${\rm y}=\tau$.
217: With this interpretation, eq.~(\ref{RGEA}) describes effectively a
218: 2+1 dimensional field theory (the two transverse coordinates plus the
219: evolution time), which is however {\it non-local} in
220: both $x_\perp$ and ${\rm y}$
221: (since Wilson lines likes (\ref{vtau}) involve integrals over all
222: ${\rm y}\in (-\infty,\tau)$).
223:
224: {\sf vii) The infrared and ultraviolet behaviours of the RGE.}
225: These are determined by the corresponding
226: behaviours of the kernel $\eta^{ab}(x_\perp,y_\perp)$ (we shall
227: shortly see that $\nu$ is related to $\eta$), and, more precisely,
228: by the behaviour of the integrand in eq.~(\ref{eta}) for
229: fixed $x_\perp$ and $y_\perp$ (since these are the ``external''
230: points at which we probe correlations in the system).
231: For the infrared, we need this integrand in the
232: limit where $z_\perp$ is much larger than both $x_\perp$ and $y_\perp$.
233: Then the products of Wilson lines involving $z_\perp$ are expected to
234: be small (since, e.g.,
235: $\langle V^\dagger_x V_z \rangle \to 0$
236: as $|z_\perp-x_\perp|\to \infty$ \cite{SAT}),
237: so it is enough to study the large--$z_\perp$ behaviour of
238: \be\label{Kxyz}
239: {\cal K}(x_\perp,y_\perp,z_\perp)\,\equiv\,
240: \frac{(x^i-z^i)(y^i-z^i)}{(x_\perp-z_\perp)^2(y_\perp-z_\perp)^2}\,.\ee
241: For $z_\perp \gg x_\perp,y_\perp$, this gives
242: ${\cal K}_{IR} \sim 1/z_\perp^2$ and the ensuing integral
243: $(d^2z_\perp/z_\perp^2)$ has a logarithmic infrared divergence.
244: Thus, there is potentially an IR problem in the RGE. This is
245: not necessarily a serious difficulty, since IR problems are expected
246: to be absent only for the {\it gauge-invariant} observables.
247: In this context, we would expect the IR divergences to cancel out
248: when the RGE is used to derive evolution equations for
249: gauge-invariant observables. Although a general proof in this sense
250: is still missing, we shall nevertheless see some explicit examples
251: where such cancellations take place indeed.
252:
253: Coming now to the ultraviolet, or short-range,
254: behaviour, it is easy to see on eqs.~(\ref{eta}) or
255: (\ref{etasym}) that no UV problem is to be anticipated.
256: For instance, the would-be linear pole of ${\cal K}(x_\perp,y_\perp,z_\perp)$
257: at $|z_\perp-x_\perp|\to 0$ is actually cancelled by the
258: factor $1- V^\dagger_z V_x$ which vanishes in the same limit.
259:
260: \subsection{The RGE in Hamiltonian form}
261:
262: The RGE (\ref{RGEA}) is a functional Fokker-Planck equation,
263: that is, a diffusion equation for a (functional) probability density.
264: It portrays the quantum evolution towards small $x$ as
265: a random walk (in time $\tau$) in the functional space spanned
266: by the two-dimensional color fields $\alpha^a(x_\perp)$.
267: The term involving second order derivatives in this equation is
268: generally interpreted as the ``diffusion term'', while that
269: with a single derivative is rather the ``force term''.
270: But this interpretation is not unique for our RGE where the
271: 2-point function $\eta^{ab}(x_\perp,y_\perp)$ --- the
272: analog of the diffusion ``constant'' --- is itself a %(non-local)
273: functional of the field variable $\alpha^a(x_\perp)$, so that
274: eq.~(\ref{RGEA}) may be as well rewritten as
275: \be\labe{RGEAnew}
276: {\del W_\tau[\alpha] \over {\del \tau}}\,=\,
277: {\delta \over {\delta \alpha_\tau^a(x_\perp)}}
278: \left\{ {1 \over 2} \eta_{xy}^{ab} \,{\delta W_\tau\over
279: {\delta \alpha_\tau^b(y_\perp)}} + \left({1 \over 2} \,
280: {\delta \eta_{xy}^{ab}\over {\delta \alpha_\tau^b(y_\perp)}}
281: -\nu_{x}^a\right)W_\tau \right\}\,,
282: \ee
283: which rather features ${1 \over 2}
284: ({\delta \eta/{\delta \alpha_\tau}}) -\nu$ as the
285: effective ``force term''.
286: An important property, which is full of consequences, is that
287: this effective ``force term'' is precisely zero.
288:
289: Specifically, the following functional relation holds between
290: the coefficients of the RGE:
291: \be\label{sigchi}
292: {1 \over 2} \int d^2y_\perp
293: {\delta \eta^{ab}(x_\perp,y_\perp)\over {\delta \alpha_\tau^b(y_\perp)}}
294: \,=\,\nu^a(x_\perp)\,.\ee
295: It is straightforward to prove this relation by computing the
296: effect of $\delta/\delta \alpha_\tau^b(y_\perp)$ on the terms
297: involving Wilson lines in eq.~(\ref{eta}) for $\eta^{ab}(x_\perp,y_\perp)$.
298: For instance,
299: \be\label{anti}
300: {\delta\over {\delta \alpha_\tau^b(y_\perp)}}\,(V^\dagger_x V_y)^{ab}
301: &=& {\delta V^{\dagger\,ac}_x\over {\delta \alpha_\tau^b(y_\perp)}}\,V_y^{cb}
302: \,+\,V^{\dagger\,ac}_x{\delta V_y^{cb}\over {\delta \alpha_\tau^b(y_\perp)}}\nn
303: &=& ig\delta^{(2)}(x_{\perp}-y_\perp)\Bigl(T^bV^\dagger_y\Bigr)_{ac}V_y^{cb}
304: -ig\delta^{(2)}(0_{\perp})V^{\dagger\,ac}_x\Bigl(V_yT^b\Bigr)_{cb}\,=\,0\,,\ee
305: where each of the two terms vanishes because of the antisymmetry of the
306: color group generators in the adjoint representation (e.g.,
307: $(T^b)_{ab}=0$). One can similarly show that:
308: \be
309: -{\delta\over {\delta \alpha_\tau^b(y_\perp)}}\,(V^\dagger_x V_z)^{ab}
310: &=&-ig\delta^{(2)}(x_{\perp}-y_\perp)\Bigl(T^b V^\dagger_x V_z\Bigr)_{ab}
311: \,=\,ig\delta^{(2)}(x_{\perp}-y_\perp)
312: {\rm Tr}\Bigl(T^a V^\dagger_x V_z\Bigr),\nn
313: -{\delta\over {\delta \alpha_\tau^b(y_\perp)}}\,(V^\dagger_z V_y)^{ab}&=&0.\ee
314: The only non-vanishing contribution is that in the first line above,
315: and this precisely reproduces eq.~(\ref{nu}) after integration over $y_\perp$,
316: since (cf. eq.~(\ref{Kxyz}))
317: \be\label{Kxxz}
318: {\cal K}(x_\perp,x_\perp,z_\perp)\,=\,\frac{1}{(x_\perp-z_\perp)^2}\,.\ee
319: By using eqs.~(\ref{RGEAnew}) and (\ref{sigchi}), the RGE
320: can be finally brought into the Hamiltonian form:
321: \be\labe{RGEH}
322: {\del W_\tau[\alpha] \over {\del \tau}}\,=\,-\,H W_\tau[\alpha],
323: \ee
324: with the following Hamiltonian:
325: \be\labe{H}
326: H&\equiv&
327: {1 \over 2}
328: \int d^2x_\perp\int d^2y_\perp\,
329: {\delta \over {\delta
330: \alpha_\tau^a(x_{\perp})} }\,\eta_{xy}^{ab}\,
331: {\delta \over {\delta \alpha_\tau^b(y_{\perp})}}\,=\,
332: \int {d^2 z_\perp\over 2\pi }\,J^i_a(z_\perp)\,J^i_a(z_\perp),\nn
333: J^i_a(z_\perp)&\equiv& \int {d^2 x_\perp\over 2\pi }\,
334: \frac{z^i-x^i}{(z_\perp-x_\perp)^2}\,(1 - V^\dagger_zV_x)_{ab}\,
335: {i \delta \over {\delta
336: \alpha_\tau^b(x_\perp)} }\,,\ee
337: which is Hermitian (since $\eta_{xy}^{ab}$ is real and symmetric)
338: and positive semi-definite (since the ``current'' $J^i_a(z_\perp)$
339: is itself Hermitian).
340:
341: At a first sight, the interpretation of eqs.~(\ref{RGEH})--(\ref{H})
342: as a Hamiltonian system may look compromised by the fact that
343: the operator (\ref{H}) appears to be time-dependent, e.g.,
344: via the $\tau$-dependence of the differentiation point
345: $\alpha_\tau^a(x_\perp)$, and even non-local in time,
346: via the Wilson lines $V$ and $V^\dagger$ (which involve integrals
347: over all times ${\rm y}\le\tau$; cf. the discussion
348: after eq.~(\ref{alphatau})). But this rather means that we have not
349: yet identified the correct canonical variables: these are not the
350: fields $\alpha_\tau^a$, but rather the Wilson lines
351: ending at $\tau$. Specifically, the
352: canonically conjugate variables and momenta are
353: $\{U^{ab}_\tau(x_{\perp}),\,\Pi^c_\tau(y_\perp)\}$, with
354: $U_\tau(x_{\perp})$ as defined in eq.~(\ref{vtau}), and
355: $\Pi^a_\tau(x_\perp)$ acting on $U_\tau$ or $U^\dagger_\tau$
356: as the functional derivative w.r.t. the field at the end point:
357: \be\labe{CANPI}
358: \Pi^a_\tau(x_\perp)\,\equiv\,\frac{1}{ig}\,{ \delta \over {\delta
359: \alpha_\tau^a(x_\perp)} }\,.\ee
360: The associated Poisson brackets, or equal-time commutation
361: relations, are easily inferred from eq.~(\ref{DIFFU}):
362: \be\label{PB}
363: \Big[\Pi^a_\tau(x_\perp),\,U^\dagger_\tau(y_{\perp})\Big]&=&
364: T^a U_\tau^\dagger (y_{\perp})\delta^{(2)}(x_{\perp}-y_\perp),\quad
365: \Big[\Pi^a_\tau(x_\perp),\,U_\tau(y_{\perp})\Big]\,=\,-
366: U_\tau(y_{\perp})T^a\delta^{(2)}(x_{\perp}-y_\perp),\nn
367: \Big[\Pi^a_\tau(x_\perp),\,\Pi^b_\tau(y_{\perp})\Big]&=&
368: if^{abc}\Pi^c_\tau(x_\perp)\delta^{(2)}(x_{\perp}-y_\perp),
369: \qquad\Big[U_\tau^{ab}(x_{\perp}),\,U_\tau^{cd}(y_{\perp})\Big]
370: \,=\,0.\ee
371: (Non-shown commutators like $[ U_\tau^\dagger, U_\tau^\dagger]$
372: or $[ U_\tau^\dagger,U_\tau]$ vanish as well. This should not be
373: confused with the fact that, as color matrices, $U_\tau(x_{\perp})$
374: and $U_\tau(y_{\perp})$ do not commute with each other at different
375: points $x_{\perp}\ne y_{\perp}$.)
376:
377: Eqs.~(\ref{vtau}) and (\ref{CANPI}) provide explicit
378: representations for the canonical variables in terms
379: of the gauge field $\alpha_{\rm y}^a(x_\perp)$. But we can extend
380: these relations to more abstract definitions, which hold
381: independently of any representation.
382: The generic canonical ``coordinates'' are
383: the group-valued 2-dimensional fields $V^{ab}(x_{\perp})$, while
384: the canonical ``momenta'' $\Pi^a(x_\perp)$ are Lie derivatives
385: w.r.t. these fields, whose action is {\it defined} by eqs.~(\ref{PB}).
386: These momenta satisfy the commutation relation of the Lie algebra,
387: as they should (cf. the second line of eq.~(\ref{PB})).
388: In terms of these variables, the Hamiltonian (\ref{H}) takes the more
389: standard form of a second-quantized Hamiltonian for a 2-dimensional
390: field theory:
391: \be\labe{H2q}
392: H[\Pi,V]\,=\,
393: {1 \over 2}
394: \int d^2x_\perp\int d^2y_\perp\,\Pi^a(x_\perp)\,\Big(g^2
395: \eta_{xy}^{ab}[V,V^\dagger]\Big)\,\Pi^b(y_\perp)\,.\ee
396: Note that this is a purely kinetic Hamiltonian (no potential).
397: It describes free motion on a non-trivial manifold (here, a
398: functional group manifold),
399: with the kernel $g^2\eta^{ab}_{xy}[V,V^\dagger]$ playing the role
400: of the metric on that manifold. Then, (\ref{RGEH}) is like
401: the evolution equation for this Hamiltonian in imaginary time.
402: Actually, giving the probabilistic interpretation of $W_\tau[\alpha]$,
403: the best analogy is not with the (functional) Schr\"odinger equation
404: in imaginary time, but rather with the Fokker-Planck equation.
405: We shall return to this analogy by the end of this section (see also
406: \cite{BIW}).
407:
408:
409: The Hamiltonian (\ref{H2q}) has been first obtained by Weigert
410: \cite{W} in a rather different context, namely from an analysis
411: of Balitsky's equations. These are coupled evolution equations
412: for Wilson-line operators that have been derived by Balitsky \cite{B}
413: via the operator product expansion of high-energy QCD scattering
414: in the target rest frame.
415: Weigert has subsequently recognized that all these equations,
416: which form an infinite hierarchy, can be generated from a
417: rather simple functional evolution equation with the Hamiltonian
418: (\ref{H2q}). The fact that we have come across the same Hamiltonian means
419: that our RGE too generates Balitsky's equations,
420: which we shall verify explicitly in the next section on the
421: example of the 2-point function
422: $\langle {\rm tr}\big(V^\dagger(x_{\perp}) V(y_{\perp})\big)\rangle_\tau$.
423:
424: %The representation (\ref{H2q}) is especially useful if one
425: %is interested just in Green's functions of the Wilson lines, like
426: %$\langle {\rm tr}\big(V^\dagger(x_{\perp}) V(y_{\perp})\big)\rangle_\tau$.
427: %Given $W_\tau[\alpha]$, the probability density for all such
428: %correlations is formally given by
429: %\be\label{U-alpha}
430: %Z_\tau[V,V^\dagger]\,=\,\int {\cal D}\alpha\,\,\delta\Big(V^\dagger(x_\perp)
431: %- {\rm P} {\rm e}^{ig \int_{0}^{\infty} dx^-\,\alpha (x^-,x_{\perp})
432: % }\Big)\,W_\tau[\alpha]\,,\ee
433: %where the $\delta$-function restricts the functional
434: %integral to the
435: %``trajectories'' $\{\alpha_\eta^a(x_\perp)\,|\,0\le x^- <\infty\}$
436: %leading to a given group element $V^\dagger$.
437:
438: Weigert has been also the first one to notice the remarkable
439: relation (\ref{sigchi}) between the coefficients in the functional
440: evolution equation. Let us devote the end of this section to a more
441: thourough discussion
442: of this relation and some of its consequences.
443:
444: As a relation between one- and two-point functions, eq.~(\ref{sigchi})
445: is reminiscent of a Ward identity, and is most probably a consequence
446: of gauge symmetry, although we have not been able to demonstrate
447: this convincingly. It can be checked that a corresponding relation holds
448: between the coefficients $\tilde\chi$ and $\tilde\sigma$
449: in the RGE for $W_\tau[\tilde\rho]$ :
450: \be\label{sigchiCOVG}
451: {1 \over 2} \int d^2y_\perp
452: {\delta \tilde\chi^{ab}(x_\perp,y_\perp)\over
453: {\delta \tilde\rho_\tau^b(y_\perp)}}
454: \,=\,\tilde\sigma^a(x_\perp)\,\ee
455: (this is a rather obvious,
456: consequence of eqs.~(\ref{sigchi}) and (\ref{nudef})) and, more
457: significantly, also between the coefficients $\chi$ and $\sigma$
458: of the original RGE (\ref{RGE}) for $W_\tau[\rho]$ :
459: \be\label{sigchiLC}
460: {1 \over 2} \int d^2y_\perp
461: {\delta \chi^{ab}(x_\perp,y_\perp)\over {\delta \rho_\tau^b(y_\perp)}}
462: \,=\,\sigma^a(x_\perp)\,.\ee
463: This last relation is more significant in that it sheds some more light
464: on the classical polarization term $\delta\sigma$ : this is such as
465: to make the two relations (\ref{sigchiCOVG}) and (\ref{sigchiLC})
466: consistent with each other, and with the $\rho$--dependence of the
467: gauge rotation (\ref{tildechi}) from $\chi$ to $\tilde\chi$.
468: That is, $\delta\sigma$ has precisely
469: the right value to compensate the functional derivatives of
470: the Wilson lines $V$ and $V^{\dagger}$ in eq.~(\ref{tildechi}).
471:
472: There are at least two important consequences of the relation
473: (\ref{sigchi}). The first one refers to the cancellation of
474: infrared divergences in the evolution equations for gauge-invariant
475: quantities. We shall see in Sect. \ref{sec:BK} that,
476: when the Balitsky--Kovchegov equation is derived directly
477: from the RGE in Hamiltonian form (\ref{RGEH}) --- where eq.~(\ref{sigchi})
478: has been already used ---, the IR divergences are automatically absent.
479: But if one rather starts with the original RGE (\ref{RGEA}),
480: where both $\eta$ and $\nu$ are still present, then each of these
481: terms will individually give rise to IR singularities, which
482: will cancel only in the final sum. We expect similar
483: cancellations to work for all gauge-invariant observables,
484: but a general proof in this sense is still lacking. This would be
485: the non-linear generalization of the standard cancellation of
486: IR divergences between virtual and real corrections.
487:
488: The second consequence refers to the behaviour of the weight function
489: $W_\tau[\alpha]$ at large ``times'' $\tau$, which is what
490: governs the high energy limit of gluon correlations and,
491: ultimately, of hadron cross-sections. To better appreciate the connection
492: between eq.~(\ref{sigchi}) and the large time behaviour, it is useful to
493: consider first the simpler example of the Brownian motion,
494: which is governed too by a diffusion equation similar to eq.~(\ref{RGE}),
495: but for which an explicit solution can be readily worked out.
496: Consider thus a particle suspended in a highly viscous liquid,
497: and in the presence of some external force. The particle performs
498: a random walk because of its random collisions off the molecules
499: in the liquid. The relevant quantity %to characterize this
500: --- the analog\footnote{This analogy will be developed in more
501: detail somewhere else \cite{BIW}.} of $W_\tau[\alpha]$ --- is
502: $P(x,t)$ ($\equiv$ the probability density
503: to find the particle at point $x$ at time $t$), which obeys
504: the following (Fokker-Planck) equation \cite{ZJ,Parisi}:
505: \be\label{FPBM}
506: {\del P(x,t)\over {\del t}}\,=\,D{\del^2\over \del x^i\del x^i}\,P(x,t)\,-\,
507: {\del\over \del x^i}\Bigl(F^i(x) P(x,t)\Bigr),\ee
508: where $D$ is the diffusion constant (for simplicity, we assume
509: this to be truly a constant, i.e., independent of $x$ or $t$),
510: and $F^i(x)$ is the ``external force'' (actually, the force divided
511: by the viscosity).
512: If $F^i=0$, the solution is immediate, and reads (with the
513: initial condition $P(x,0)=\delta^{(3)}(x)$):
514: \be P(x,t)\,=\,{1\over (4\pi Dt)^{3/2}}\,\,
515: {\rm exp}\left\{-\frac{x^2}{4Dt}\right\}.\ee
516: This is normalized to unity, as it should, at any $t$
517: ($\int d^3x P(x,t) =1$), and goes smoothly to zero at any $x$ when
518: $t\to \infty$ (runaway solution). Thus, for sufficiently large time,
519: the probability density is quasi-homogeneous (no $x$ dependence),
520: but it is everywhere close to zero (to cope with the normalization
521: condition). The particle simply diffuses in all the available space.
522:
523: This situation changes, however, if the motion of the particle
524: is biased by an external force. Assume this force can be derived from
525: a potential:
526: \be F^i\,=\,-{\del V\over \del x^i}\,.\ee
527: Then the time-independent distribution $P_0(x)\sim
528: {\rm exp}[-\beta V(x)]$ is a stationary solution to eq.~(\ref{FPBM})
529: provided $\beta D=1$. Of course, this solution is acceptable as a probability
530: density only if it is normalizable, which puts some constraints on the form
531: of the potential. But assuming this to be the case, then $P_0(x)\sim
532: {\rm e}^{-\beta V}$ represents an equilibrium distribution which is
533: (asymptotically) reached by the system at large times. Once this is
534: done, all the correlations become independent of time.
535:
536: Returning to our quantum evolution, a stationary distribution would
537: correspond to a non-trivial fixed point of the RGE (\ref{RGE}) (a solution
538: ${\cal W}_0[\alpha]$ which is normalizable and independent of time).
539: If such a distribution existed, then the high energy limit of QCD
540: scattering would be trivial: at sufficiently high energies,
541: all the cross sections would become independent of the energy.
542: The relation (\ref{sigchi}) between the coefficients in the RGE
543: guarantees, however, that this cannot happen: The effective force
544: in eq.~(\ref{RGEAnew}) vanishes, and the corresponding evolution
545: Hamiltonian (\ref{H}) is a purely kinetic operator, which
546: describes diffusion on the group manifold.
547: We do not expect any non-trivial fixed
548: point\footnote{Of course, a constant distribution
549: ${\cal W}_0=const.$ (no dependence on $\alpha$) would be a stationary
550: point for the Hamiltonian, but this should scale as $1/V$ with $V=$
551: the volume of the manifold --- to be normalizable ---,
552: and thus would vanish as $V\to \infty$.}
553: for this Hamiltonian, just runaway solutions, and this is
554: indeed the case for the approximate solutions to eq.~(\ref{RGEH})
555: found in Ref. \cite{SAT}.
556:
557: