August 1966 1045 JOURNAL OF THE OPTICAL SOCIETY OF AMERICA VOLUME 56, NUMBER 8 AUGUST 1966 Validity of the Rytov Approximation in Optical Propagation Calculations* W. P. BROWN, JR. Hughes Research Laboratories, Malibu, California 90265 (Received 13 January 1966) The applicability of the Rytov approximation to the calculation of the characteristics of optical propagation in a weakly inhomogeneous random medium is investigated. The condition that the mean square value of the second term in the associated perturbation expansion be smaller than that of the first is adopted as a criterion for the validity of the Rytov approximation. It is shown that there is a very severe range limitation on the validity of the Rytov approximation for optical propagation in the lower portions of the earth's atmosphere. Furthermore, comparison of the limiting form for large xzkl of the validity condition derived in this paper with the condition on the validity of the Born approximation obtained independently by Mintzer, and by Kay and Silverman reveals that the Rytov and Born approximations have the same domain of validity in this limiting case. The equivalence of the validity conditions for the Rytov and Born approximations contradicts the statements of Tatarski and Chernov who contend that the Rytov approximation is superior to the Born approximation. It is conjectured that the Rytov and Born approximations have the same domain of validity for all x/klP. INDEx HEADINGS: Atmospheric optics. I. INTRODUCTION THE work reported in this paper was prompted by Tthe remarks of Hufnagel and Stanley' regarding the validity of the Rytov approximation employed in the work of Obukhov, 2 Chernov, 3 and Tatarski.A On the basis of a qualitative argument which appears to be substantiated by experimental data, they conclude that the validity of the Rytov approximation is limited in the case of optical propagation in the atmosphere. At present there appears to be no general agreement among those interested in optical propagation, concerning this * Work supported by the U. S. Air Force Office of Scientific Research, under Contract AF 49(638)-1439. ' R. E. Hufnagel and N. R. Stanley, J. Opt. Soc. Am. 54, 52 (1964). 2 A. M. Obukhov, Izvestiya Akad. Nauk., Geophys. Ser. 2, 155 (1953) [translated by WV.C. Hoffman in Project RAND Report T-47 (1955)]. 3 L. A. Chernov, Wave Propagation in a Randomt Medium (McGraw-Hill Book Co., Inc., New York, 1960). 4V. I. Tatarski, Wave Propagation in a Turbulent Medium (McGraw-Hill Book Co., Inc., New York, 1961). question. The purpose of this paper is to corroborate mathematically the qualitative conclusions of Hufnagel and Stanley. The applicability of the Rytov approximation to the calculation of the characteristics of optical propagation in a weakly inhomogeneous atmosphere is investigated. The point of departure for the calculations is the observation that the wave-fluctuation 'approximations derived by the method of Rytov are simply the first terms in perturbation expansions valid for sufficiently weak atmospheric inhomogeneities. The condition that the mean square value of the second term in the perturbation expansion be smaller than that of the first is adopted as a criterion for the validity of these results. Violation of this condition implies that the perturbation expansion is not uniformly valid and thus that the approximation cannot be improved by including additional terms in the expansion. Experience with similar problems suggests that the nonuniformity of the perturbation expansion occurs at large propagation distances. This prediction is verified by the calculations. W. P. BROWN, 1046 JR. Vol. 56 Since this discussion is concerned primarily with the validity of the Rytov approximation as applied by Chernov and Tatarski, we have attempted to follow their analysis closely to facilitate a comparison. In order to obtain a quantitative estimate of the mean square value of the second term in the perturbation expansion, however, it is necessary to assume that refrac- the imaginary part to the perturbation in the wavefront (phase). Substitution of (2) in (1) yields the following nonlinear equation for it'(r): tive index of the atmosphere is a homogeneous isotropic Rytov obtains an approximate solution of (3) by assuming that the terms V~q(r)* Vi,1(r) and k262n512(r) can be neglected. According to Rytov (and those who have adopted his approach), the legitimacy of this assumption is guaranteed if the following conditions are satisfied: gaussian random process. This assumption is more restrictive than that of Tatarski, who assumes only that the refractive index is locally homogeneous and isotropic. There is good reason to believe, however, that the limitation on the applicability of the Rytov approximation derived in this paper also applies for more general random media. The nonuniformity of the perturbation expansion is primarily a defect associated with the method of approximation and, as such, should not depend critically on the details of the statistical characteristics of the medium. V2Pj(r)+ V~p(r) *E2k 2,6o(r)+n7,Pj(r)0 + 2k 260(r)nj1(r) + *232n12(r)= 0. 5<<1i, (3) \ (4) I Vb I1[<<k. (5) The calculations given in this paper reveal that (4) and (5) are not sufficient conditions to ensure the validity of the approximation obtained when the abovementioned terms are neglected. II. METHOD OF RYTOV-PERTURBATION EXPANSION B. Perturbation Expansion A. Formulation of the Problem Suppose that the change of the refractive index in the time required for propagation is negligibly small. Then, under the additional hypothesis that the change of the index over a distance of one wavelength Implicit in the approach of Chernov and Tatarski is the assumption that Al1(r)possesses an asymptotic expansion in powers of the small parameter is small, the 00 characteristics of electromagnetic waves propagating in an inhomogeneous medium can be derived from the solution of the reduced wave equation, 3: #z(r)= B E 6j 'j(r). (6) 5 [V 2 +k 2 n 2 (r)]U(r) = 0, When (6) is substituted in (3) and the coefficients of the (1) where k is the wavenumber =27r/N, n(r) is the refractive index, and the harmonic time dependence eriwt has been assumed. In the case of wave propagation various powers of tion of (1) is written in the form U(r)= exp[#o(r)+#i(r), 2k24n(r)n 1 (r), Xo(r) = - Xl(r) = - k 2 nl12(r) - Vtlo* VV'io, (8) (9) In-1 X.(r) =- Z v~Pj,(r)- vq/i,0 0 psj(r), On, 2. (10) p==O (2) To complete the specification of the slj(r), conditions with Vo(r) chosen such that expE[o(r)] is the solution (1) in the limit of zero S (i.e., no fluctuations). (7) where sents the mean value of the refractive index, 3 is a di- mensionless parameter much less than unity, and nl(r) is a random function of position that accounts for the fluctuations in refractive index. In accordance with the method of Rytov, the solu- are equated to zero, a recursive sys- (V2 +2Vipo-V)tfrim(r)= X.(r), in a turbulent atmosphere, the refractive index is composed of the sum of two terms, n(r) =no(r)+bni(r). The term no(r) is a sure function (nonrandom) which repre- a tem of partial differential equations is obtained for the &ij(r).The rnth term in (6) satisfies the equation It is apparent that the real part of 4+(r)is related to the perturbation in the amplitude of U(r) for nonzero a and I W. C. Hoffman, IRE Trans. Antennas Propagation Special Suppl., S301 (1959). must be assigned at the boundaries of the domain of inhomogeneity. Plane-wave propagation in the x direction is considered and, in conformity with the assumptions of Chernov and Tatarski, the tli(r) are required to remain bounded when x t-> Wc. Equation (7) is solved for the case of propagation in a medium characterized by a refractive index that has a mean value of unity, no(r)= 1, and a normalized random VALIDITY August 1966 OF THE RYTOV APPROXIMATION component given by It is assumed thati/jr) xl(r) = g(x)vi(r). (11) If Vlj(r= The quantity g(x) is a sure function of the coordinate along the initial direction of propagation. The functional form of g(x) is assumed to be such that the integrals which appear in the solution of (7) remain bounded. This factor is included in anticipation of the nonuniformity of (6) for propagation in an unbounded inhomogeneous medium. The function P1(r) is taken to be a homogeneous isotropic random field; as such, it has the two-dimensional Fourier-Stieltjes spectral representation.6 vi(r) = ff exp[i. *rtd1dN(xx), (16) where dkj(vi,x) is a random process determined by the equation that results when (16) is substituted in (7). Under this hypothesis, the source function Xj(r) in (7) can also be represented by a two-dimensional FourierStieltjes integral Xj(r)= f exp[ii. rt]dXj(ux). (17) The quantity dXj(ti,x) is a random process determined, in general, by dN(;,x) and a convolution of the e^,,+ d*'rmcx), with m< j. e zuz, (13) (14) The quantity dN(c,x) is a random process that satisfies the relation (dN(Icx)dN*(,cx/))av = (ic-)/ic, X-X' I )dndic', (15) Substitution of (16) and (17) in (7) yields the following differential equation for d4't(x,x) (note that to(r)=ikx for the case of plane wave propagation in a medium where no(r) = 1): (allaxl+i2k8ldx-KI)d~j(scx)=dX 1(E,x) (18) with Ki2=x . The general solution of (18) is easily obtained by variation of parameters. When the boundedness condition for x oo is invoked, the solution becomes where the ( ),vdenotes a statistical average, 3(c- vC')is a two-dimensional delta function, and F,(jrx-x'j) is the two-dimensional spectral density of Pi(r). dx' exp{i[k- (k2 -K2 ) ]x'}dXj(ic,x') (exp-(i(k-(k2-K2)]x}f +exp- {ilkj+ (k2-K2)2]X}f The quantity dXo(cx) exp[i.-rjdtj(t,x), (12) rt= eyy+ezz. dtj(cx)=- has the spectral representation -00 with (ei= unit vector in ith direction) 'K = 1047 dx' exp{i[k+ (k2- K2)2]x'}dXj(1C,x')). (19) is easily obtained from (8), (11), (12), and (17): dXo(x,x) = - 2k2 g(x)dN(x,x). (20) Substitution of (20) in (19) then yields an expressionfor the two-dimensional Fourier-Stieltjes transform of the first term in the perturbation expansion (6). d4ro(xx)= ik2 (k2 - / exp-(i[k-(k 2 X - 2 K) ]x}j dx' exp{i~k-(k 2 -K2) -K x'}g(x')dX(cx') K~2) +exp- (i~k+(k 2- K2) ]X}f 2 dx' exp{i[k+ (k2-Ki2)]x')g(x')dN (i,x')). (21) Equation (21) reduces to the result of Tatarski 7 if it is assumed that g(x') is unity in the interval 0<x'<x and goes smoothly to zero outside this interval (approximately zero within a few correlation lengths from the end points x'=0, x). In this case, it can be shown that the second integral in (21) is negligible and the first can be approxi- 6A. M. Yaglom, An Introduction to threTheory of Stationary Random Functions, R. A. Silverman, Ed. (Prentice Hall, Englewood Cliffs, N. J., 1962). V. I. Tatarski, Ref. 4, pp. 128-130. 4 A X n WV. P. 1(148 BROWN, mated by replacing (k2-K2)' with k- ( 2/2k), and the limits dio(Ex)-ik exp- [-x] f JR. Vol. 56 to x by 0 to X. Thus -C dx' exp+[i-x ]dX(x,'). (22) Similarly, a relatively straightforward calculation reveals that under the same assumptions the two-dimensional Fourier-Stieltjes transform of the second term in the perturbation expansion is approximately ik z K2 K +-exp-pFt] 2k L2kJ 7dxyexp-X [•x']2 ff ( 'dPo(K-itr)dto(4r') (23) with d'Fo(cx) given by (22). Tatarski8 has calculated the mean square value of the real and imaginary parts of the approximation to 4u10(r) obtained from (22). Addition of his results yields an expression for the mean square value of #1 0(r). (4lo(r)VPro*(r))vz..2k2xff duf duE4(x,u). (24) An estimate of the mean square value of #1 1(r) is derived in the following section. It is shown later that 82(#ll(r)&n*(r))av is comparable in magnitude to (iio(rAQio*(r))Xv when x is sufficiently large. III. ESTIMATION OF THE MEAN SQUARE VALUE OF THE SECOND TERM IN THE PERTURBATION EXPANSION In this section the mean square value of 4'1(r) is estimated for the case in which (1) Pi(r) is a gaussian random process, and (2) the second-order moments of vP(r) are gaussian. The first assumption permits expression of the fourth-order moments of the refractive-index fluctuations in terms of a permutation of the products of the secondorder moments. When the relation between the second-order moments and the two-dimensional spectral density of vi(r) is taken into account, it is found that the fourth-order moments have the following representation 4 (TI dN(ici,Xi))av= { (Xl+ i=1 x2)8(K 3 +K 4)F,(l1,Xl2)Fv(t 3 ,Xa34)+8(xi+ a3)6(c2+K4)F,(c1,Xl3)Fv(x 2 ,X7l 24 ) + 6(x±1+ C4) 6c(1 2 + x 3 )F,(61,X1 4)Fd(i 2 ,x 2 3)} dYc1dX2 dc3dX4 , (25) where xij = x-jx . The second assumption is made to simplify the task of evaluating the integrals that appear in (#fr)1l*(r)Xav. With this assumption, the two-dimensional spectral density of the ijth moment of vi(r) canrbe written 2 Fp(xi,xi;) = (12 /4,r) exp[-Kc 2 12/4-Xij21 /] 1 (26) where I is the correlation distance of the fluctuations. A formal representation of the mean square value of i/l'(r) in terms of the two-dimensional Fourier-Stieltjes transform dTI(i,x) is obtained when (16) (with j= 1) is multiplied by its complex conjugate and a statistical average taken. ff11 x00 exp[i(it- ) r,](dtl(;cx)d *(ca,,x))av. (27) In the case of interest here, i.e., long distance propagation, the contribution of the first term in (23) to (#611(r)46l1*(r)).v is negligible in comparison with that of the second term. 8 See Ref. 4, pp. 130-137. VALIDITY August 1966 OF THE RYTOV APPROXIMATION 1 049 Thus, the statistical average (dt(xcx)dt*(c 0 x)) 5 7 is approximated by [ (K2-Ka2)X] j 5 dx'f dx" exp~fXK 1 (dtl(cx)d4l*(Vcax))av~ -e x ff ff (xC Ka,2Xf/ " Y.'()a I">rc"(d4o(c-Yc',xO)dto(cXf)dta*(,a K') -xc K2XI- -c", x")dro(i",x"))>av. (28) -cc The dependence of (tn(r)ta*(r))av on the spectral characteristics of the refractive-index fluctuations is made explicit when the results expressed in (22), (25), (27), and (28) are combined. Upon rearrangement of the resultant expression, we find that ( k- f2 dKii / (r) di4 j 4 ii d pX dx'f j indx"I px' if"dA/ dvii dv2I Jo Jo Jo -00 -0: (C' ma" dv4(MiA+M2), (29) Jo where M 1 and M 2 denote the functions 2 M1 = K12K2 exp(i/2k(-K2 (Vl+v2)]+K22[2x"- [2x'- (V3+ V4)]})F6(itVl2)F,(ic2,v3 2 M2I=2 (iKi 1cK2) exp{i/2k[2xi: 1c2(x -X")+Kl'(Vl-V3)+K2 2 4) (V2-v 4)]}F96(c,vn3)Fv(62,v2 4) with vpj= jvi-vjI. A. Estimation of the Integrals Involving Ml The rather formidable task of evaluating the integrals in (29) is greatly simplified by the gaussian correlation assumption (2) and because the present paper is concerned with the value of (29) when x>>d,k>>1. Consider first the term involving Ml. Note that Ml is the product of a function of Kci,x', VI, V2and the complex conjugate of the same function with il, x', VI, V2replaced by x2, x", V3, V4.Thus, the contribution of Ml to (4ln(r)itc*(r))av is given by (k2/4)LIL 1 * where L 1 is the integral L1=, J d.cij drif dx'j dV2 21iexp{ - (iKc12 2k)[2x'- (V1+V 2 )]}F,(c1i,v 12). (30) When the variables v,, V2in (30) are replaced by new variables u= v 1 -v 2 , W=Vl+V2,the resultant integral with respect to w is easily evaluated. This yields 1=-i2k l d. dx' duF,(l A)j[.exp- Jo o [K1 (u- 2x')-i i itiit] xpi 2 2k . (31) From (26) it is apparent that F,(Ei,u) is very nearly zero for all values of Kiand u outside the domain ci(, I-1, u l. Consequently, the exponentials exp[r4iKl2u/2k] in (31) can be replaced by unity since K1 2u/k( 1/ki<<1 in the domain where F,(xi,u) is appreciably different from zero. i - i2kJf dcij dx'[t- exp- (i )] j duFci,u). (32) Finally, note that the bracketed quantity in (32) is very small unless x't hi2 . Consequently, in view of the fact that F,(Ri,u) rapidly approaches zero when u> l,it is permissible to replace the upper limit in the integral over it by infinity. When this is done and F/(ci,u) is expressed in the form (26),9 the evaluation of the integrals in (32) is relatively straightforward. The result is L,--i(r)kxl[1+i(k1 2 /4x) ln(1+i4x/kh 2)]. (33) Actually, the assumption of gaussian correlation implicit in the use of (26) is not required in the evaluation of (32). It is possible to proceed in the manner of Tatarski, i.e., let Io F,(xl,u)du= r4(KO), where k is the three-dimensional spectral density of the refractive index fluctuations. In this case q(K,O)need not correspond to a gaussian correlation function. Equation (26) is employed in the calculation of L, however, to conform with the approach required in the evaluation of the integrals involving M2 . W. P. BROWN, 1050 JR. Vol. 56 Thus, the contribution of Ml to (qt1(r)qt,,*(r))av is (k2l4)LL,*- (.7r/4)(kl)1y2 T(y), (34a) 2 where -yis the dimensionless variable x/k1 and T(,y) = [1- (1/4,y) tan7'4y] 2 + (1/64-y2)[ln(1+ 16y 2 )] 2 . (34b) B. Estimation of the Integrals Involving M2 Integration over the variables vim,V2, v3, and V4 is facilitated by the introduction of new variables Ui=vi-vi+2, is written in the form (26) and it is 2 (i= 1, 2). When the two-dimensional spectral density F,(xifii) again noted that exp[iKti2i/2kJ- 1 in the region where the integral F, is appreciably different from zero, the integrals over ui, wi are easily evaluated. The result can be written in the form Wi= Vi+Vi+ dvlJ dV2J dvJ dV4M2-(X1 lc2)2W(X1',X',1)exp(-E(K,2+K22)12 14]+iE'cl-21k](xl'-x")), (35) where 18 W(x',x",l)=-'1+exp327r2 -(X I'-Xf/)2_ exp-- 1 X12\ -exp- X12\ 21 X- - 2- fx'/e l eU2 dite"/- I Jo1J X" 2- e-2du±X'/J -Udu+ I I 1 X'-_X'[/ e--'du. JoI le2 (36) Next substitute (35) in (29), interchange the order of integration with respect to the dus and x', x", and express the dxi in the polar form dxi= KidKid6i. After evaluating the integrals with respect to the angles si, it is found that the contribution of M2 to (ftl(r)+tl*(r))av can be expressed in the form 2k2 k2 -4LLl*=4 where W(x',x",l) 2 x dx'J dx"V(x',",1)Q(xx (37) ,1), is the function defined in (36) and Q(x',x",l) is Q8xx",l)= - 47r2 dxj f dK2{KiK2Jo[EKK2(x'-x")/k] The integral over K2 in (38) is the zero-order Hankel transform of the function K2' exp(-K 2 2 12/4). Substitution of the tabulated value of this transforms in (38), integration of the result over Ki, and differentiation with respect to x'-x" as indicated in (38) yields the following explicit representation of Q. Q(x',x",l)= (272/18)[1- 121 2 )/(1n2 +l);'1 (39) where X denotes the dimensionless variable (x'- x")/k12 . Finally, note that in the major portion of the domain of integration in (37), W(x',x",l) can be approximated by the function obtained when the upper limits of the integrals in (36) are replaced by infinity. When written in terms of the variable i defined above and a variable v= (x'+x")/k1 2 , the approximation assumes the form W(x',x",l)'-(18/128r)(kl)2 (u- 2v)2. (40) The portions of the domain of integration in (37) where (40) is not valid are confined to the immediate vicinity 10W. Magnus and F. Oberhettinger, Formulas and Tlheoremsfor the Functions of Mathlematical Physics (Chelsea Publishing Co., New York, 1954), p. 137. 2 2 expE-(K )1 _2/ 4]} (38) of the lines x'= 0, x"= 0, x'= x". Since these regions are no more than a few correlation lengths in width, the error incurred in approximating W(x',x",l) everywhere by (40) is negligible. Substitution of (39) and (40) Jin (37) and evaluation of the resultant integral yields~the following expression for the contribution of M2 to (tiu(r)tlu*(r))av, (tlu(r)ipln*(r))av,-k2/4LLi*= [Er(kl)6 /256]P(y), (41) where Oyagain denotes the dimensionless variable x/k12 and P(-y) = 1 9 2-y 2 - 144-ytan'2-y+ 28 ln(4-y 2 + 1)- 140/3 49/3 128/3 y4 + 6,y 2- 7/6 ± -- +(,Y,+ 4') (42) (0Y2 4 IV. A CRITERION FOR THE VALIDITY OF THE RYTOV APPROXIMATION As mentioned in Sec. II, the conditions 5K<1, IV/,1o(r)I<<1 are not sufficient to guarantee the validity VALIDITY OF THE RYTOV APPROXIMATION August 1966 of P40(r)as an approximation to i64(r). A more basic re- quirement is that the perturbation expansion (6) asymptotically represent #l(r). An asymptotic expansion is characterized by the fact that each successive term in the expansion is smaller than the preceding term for sufficiently small values of the perturbation parameter." In view of this, and since the various terms in (6) are random functions which are describable only in statistical terms, it is assumed that the validity of the expansion is related to the relative magnitude of the mean square value of successive terms. In particular, the condition that 62 times the mean square value of i',i(r) is less than the mean square value of i',o(r) is adopted as a criterion for the validity of the Rytov approximation technique: 31(,P11r+li r>a (43) (r)).v<v s(r)4,11* In effect, this condition imposes a limitation on the acceptable values of the parameters x/l and ki for a given value of 6. When (43) is not satisfied, an entirely new approach to the approximation problem is required. An integral representation of the mean square value of 4'10(r) in terms of the two-dimensional spectral density of the refractive index fluctuations is given in (24). Substitution of the gaussian spectral-density function employed in the calculations of Sec. III in (24) and evaluation of the resultant integrals yields assume that the nonuniformity of the approximation techniques (Rytov or Born) is independent of the details of the physical configuration considered and of the type of statistics which describe the medium. The equivalence of the validity conditions for the Rytov and Born approximations contradicts the statements of Tatarski,' 4 Chernov,"1 and, more recently, de Wolf,'6 who contend that the Rytov approximation has a larger domain of validity than the Born approximation. The results derived in this paper prove definitely that this is not the case when y>>4. Furthermore, the fact that both approximations suffer from the same type of nonuniformity leads to the conjecture that the Rytov and Born approximations have the same domain of validity for all y. The Rytov and Born approximations are employed in situations where the geometrical-optics approximation is invalid. Thus, to place the condition on the Rytov approximation in proper perspective, it is worthwhile to derive a condition on the validity of geometrical optics consistent with the criterion for validity adopted for the Rytov approximation. This criterion is stated in (43). The mean square value of the first term in the geometrical-optics approximation is related to the phase fluctuations and that of the second to the amplitude fluctuations. Tatarski shows that these quantities are given by the expressions' (44) (,r)1(k1)1,y. for the mean square value of #,d(r) in (43) gives the fol(w)'5%1(kl)Vy[T(y)+(1/256)PQ'6/1 < 1- (45) The significance of (45) is discussed in the following section. V. DISCUSSION AND SUMMARY Note first that when y>>1,the bracketed quantity in (45) approaches unity. In this case, the validity criterion is identical, to within a constant, with the condition on the Born approximation derived independently by Mintzer'2 and by Kay and Silverman.'3 (46) x< [(7r) 2162]-l. Mintzer considered the propagation of waves excited by a point source in a three-dimensional random medium whereas Kay and Silverman considered the propagation of a plane wave through a random stack of one- dimensional dielectric slabs. In view of the equality of the validity conditions for these diverse problems and the problem considered in this paper it is reasonable to 11A. ErdElyi, Asymptotic Expansions (Dover Publications, Inc., New York, 1956), pp. 8-13. 12D. Mintzer, J. Acoust. Soc. Am. 25, 1107 (1953). "1I. Kay and R. A. Silverman, Nuovo Cimento Suppl. 9 (Ser. 10) 626 (1958). (7r)tl 2 k2 xl, (47) (X )av= (8/3 ) (,or)132X3/13. (48) (S'2).= Substitution of this result and those derived in Sec. III lowing condition on the parameters 6, hi, and -y(= x/kl 2 ). 1051 Substitution of (47) and (48) in the validity criterion (X2)av< (Si2 1)., and simplification of the result yields the condition x<0.61k1', (i.e., y<0.61). (49) This condition agrees with the result obtained by Tatarskill on the basis of a physical argument. In a given situation, comparison of the limitations imposed on x by the conditions (45) and (49) indicates the utility of employing the more complicated Rytov approximation in place of the relatively simple geomnetrical-optics approximation. The nature of the limitation placed on the applicability of the Rytov (or Born) approximation by the inequality (45) is illustrated by Fig. 1. The three curves shown in this figure are obtained by solving equation (45) with an equality sign. The domain of validity at a given wavelength includes all points below the correspondingly labeled curve. These curves were computed for a correlation distance of 1.0 cm. Such a dis- tance is typical of the inner scale for sea level atmos14V. I. Tatarski, Ref. 4, pp. 122-126. 1"A. M. Obukhov, Ref. 2, pp. 58-66. 16D. A. de Wolf, J. Opt. Soc. Am. 55, 812 (1965). '7 V. I. Tatarski, Ref. 4, p. 148. 18V. I. Tatarski, Ref. 4, pp. 120-121. W. P. BROWN, 1052 JR. Vol. 56 classified as very weak turbulence, when SX10'-1 to relatively strong turbulence,2 0 when 3X10--10. It is apparent that the limitation on the Rytov (or Born) approximation can be rather severe in the case of propa- gation over horizontal paths at sea level. In such cases, very little is gained in going to the Rytov approximation when the light is in the red or shorter wavelength 102 portions of the spectrum. As seen from Fig. 1, the do- main of validity of the Rytov and geometrical-optics approximations 6328 lo0 A 4880 For example, when OPTICS DOMAIN OF VALtDITYFOR GEOMETRICAL lo-' 0 I 1 2 3 coalesce at 6X 10i-5.2 for 6328-A radiation and at smaller values of a for the shorterwavelength radiation. Furthermore, in this portion of the spectrum neither the Rytov nor the geometricaloptics approximation is valid at the distances currently envisaged for coherent optical propagation experiments. lo, 4 5 6I 7 8 9 1I 10 a x lo,, FIG. 1. Limiting curves for the domains of vadiditv. l= 1 cm, N=4880 1, 6328 i, and 3.39 M. pheric turbulence.'2 Generally, the domains of validity shrink as I is increased (when y>>A, the maximum range is proportional to l-l). Also, it is noted from (45) that for a given y, the maximum allowable 6 is proportional to Xd.Thus, the limiting curve for the domain of validity at a wavelength not depicted in Fig. 1 can be obtained by multiplying the values of 6 on one of the given curves by the three-halves power of the ratio of wavelengths. Finally, the maximum propagation distance over which the Rytov approximation is valid for a given degree of turbulence (i.e., a given 6) can be obtained from Fig. 1 2 . by multiplying the maximum allowable -y by Wd Available data on refractive-index fluctuations indicate that 6 is typically in the range 1<6X108<10 for sea-level atmospheric turbulence. The turbulence that causes these fluctuations ranges from what may be '"R. E. Hufnagel and N. R. Stanley, J. Opt. Soc. Am. 54, 57 (1964;. AX108 is 2, the maximum allowable Since corange at 6328 A is 2.2 km (xmax=ky12,nax). herent optical propagation experiments over horizontal 16-km paths have already been conducted,2' it is apparent that there is a need for a new theory of shortwavelength propagation valid at these distances. The situation is somewhat different in the infrared portion of the spectrum, where it appears that existing theory is adequate for most practical situations (at ranges less than a few kilometers). In summary, it has been shown that when 7y>>Z,the Rytov and Born approximations have the same domain of validity and it has been conjectured that the same is true for all 'y. In addition, it has been shown that there is a very severe range limitation on the validity of the existing theories of propagation when X< 0.6 y. The limitation in the infrared portion of the spectrum is not nearly so severe, but even here it is necessary to exclude situations in which the range is greater than a few kilometers. 20 Generally, the strength of the turbulence decreases as the altitude above the earth is increased. The data tabulated by Hufnagel and Stanley' indicate that the magnitude of the fluctuation is decreased by a factor of 10 at 3 km. 21F. E. Goodwin, paper presented at the conference on Atmospheric Limitations on Optical Propagation, Boulder, Colorado (1965). Martin Shenker (left) and L. F. Barcus, both of Farrand Optical Co., at Rochester Section's 50th Anniversary Symposium on the Use of Very Large Computers in Lens Design, 16 November 1965.