624 J. Phys. Chem.

1981, 85, 624-628

Parabolic Tunneling Calculations

Rex T. Skodje and Donald G. Truhlar*

Department of Chemistry and Chemical Physics Program, University of Minnesota, Minneapolis, Minnesota 55455
(Received: December 31, 1980)

A new analytic approximation is presented for thermally averaged transmission coefficients for tunneling through
and reflection by truncated parabolic potential barriers. The approximation has a wide range of validity and
avoids the spurious singularities of the well-known result for untruncated parabolic barriers. We also apply
the analytic result to effective parabolic fits to other barrier shapes. The results are quite encouraging and
suggest that the approximation may be useful for a wide variety of chemical applications.

I. Introduction where
Tunneling through potential energy barriers occurs in a =
2 /( )
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

many contexts in chemistry. The simplest realistic ap-

Downloaded via INDIAN INST OF TECHNOLOGY DELHI on April 4, 2019 at 13:11:16 (UTC).

proximation to a potential energy barrier is a truncated It is convenient to divide the integral into two terms
parabola. Thermally averaged transmission coefficients I 7(0, V0) + /(Vo,00) =
(7)
for truncated parabolas have usually been approximated
where
by the method of Wigner1 or that of Bell.2 In this article
we present a more satisfactory solution to this problem. tEi
exp (-ß )
We use this solution to calculate approximate transmission /(Ex,E2) =
dE (8)
coefficients for more general barrier shapes and compare *
1 + exp[a(V0 £)] -

the results to accurate uniform semiclassical (i.e., unifor- Consider first 7(0, V0). Here E < V0,
mized WKB) or quantal calculations.
so we expand the
integrand as follows
II. Theory 7(0,V0) =

A. Derivation. Symmetric Barrier. The problem under

consideration is transmission through a one-dimensional - f °expH£)
n=0
(-l)nexp[-(n + l)a(V0 -

E)] dE (9)
potential barrier. First consider a symmetric barrier
For the second term, V0 < E, and we choose a different
V(x) =
V0- y2m<b2x2 |*| < (2V0/mu2)1/2 expansion
V = 0 |*| > (2Vo/ma>2)1/2 (1)
where the particle mass and V0 and are parameters
m is
7(V0,°°) =
v
f”exp(-SE) H)n exp[na(V0 E)] -

dE
Yo n=0
of the potential. The object is to calculate a good ap- (10)
proximation to the transmission coefficient at temperature
T. This is defined for an arbitrary barrier as3 Integrating (9) and (10) and substituting into (7) and (4)
yields
Ct(E) exp (-ß ) dE
(l exp{[|8 ( + l)a] V0j
y
-

() =
(2) () =
(-!)"/
( l)a ß
-

n=0 + -

na
f
Vo
TC(E) exp(-dE) dE ()
where ß l/kBT, kB is Boltzmann’s constant, and T(E)
= When ß < a, the nonexponential parts of the sums can
and TC(E) are the quantum mechanical and classical be combined and we get
transmission coefficients at energy E. From (1) we have ß/
TC(E) =
( -

V0) (3) sin ( ß/ )

where 6(y) is a step function increasing from 0 to 1 at y (~1) ß exp{[d ~(n+ l)a]V0}
- —t:-(n-+ (ß < a)
^
=
0. Putting (3) into (2) yields =o l)a- ß
12
() =
ß exp(dVo)/ (4)
As ß approaches a from below, the pole in the first term
where of (12) is cancelled by the = 0 term in the sum. When
a is large the = 0 term dominates the sum, and
1= CT(E) exp(-dE) dE (5)
ß/
We approximate T(E) by the uniformized WKB result for () —
~r

a parabola4
sin ( ß/ )
ß
exp[(d a) V0] (aV0 » 1) (13)
-

T(E) =
(6) ß
-

a
—-7~T\y-™
1 + exp[a(V0 E)] -

For the case a =

ß this yields
(1) E. Wigner, Z. Phys. Chem. B, 19, 203 (1932).
(2) R. P. Bell, Trans. Faraday Soc., 55,1 (1959). Note that our def-
initions of a and ß are different from Bell’s. (4) E. C. Kemble, “The Fundamental Principles of Quantum Me-
(3) See, e.g., D. G. Truhlar and A. Kuppermann, J. Am. Chem. Soc., chanics with Elementary Applications”, McGraw-Hill, New York, 1937,
93, 1840 (1971). pp 109ff.

0022-3654/81/2085-0624$01.25/0 &copy; 1981 American Chemical Society

Letters The Journal of Physical Chemistry, Vol. 85, No. 6, 1981 625

() =
ß0 =
«V0 (« =
ß) (14) TABLE I: Transmission Coefficients for Symmetric
Truncated Parabolic Barriers
When « < ß, the combination of series to yield (12) is
not valid. However, for large ß one can show that2 untrun-
uni- cated Wigner
formized present para- meth-
K(T) a
exp[(|3 -

«) V0] (/3 V0 » 1) (15) V0a ß/ WKB° algorithm6 bolac odd

p
-

a
1.0 0.10 0.98 0.97 1.02 1.02
By retaining the next largest term in (11) we get 1.0 0.50 1.03 0.96 1.57 1.41
1.0 0.90 1.24 1.01 9.15 2.33
1.0 1.05 1.36 1.08 2.81
k(T) =* -r— fexp[03 a)V0] 1| (ß > a) (16)
- -

ß
-

a 1.0 1.65 2.10 2.32 5.48

1.0 2.45 4.16 5.51 10.87
Equation 16, like (13), yields the limit (14). Thus eq 13 2.0 0.10 1.00 1.00 1.02 1.02
and 16 provide a divergence-free approximation to k{T) 2.0 0.50 1.22 1.20 1.57 1.41
that is a continuous function of a and ß for all possible 2.0 0.90 1.87 1.78 9.15 2.33
values of these parameters. 2.0 1.05 2.28 2.21 2.81
B. Comparison to Previous Work. Wigner’s tunneling 2.0 1.65 5.76 6.78 5.48
correction is valid for ß « a. It can be obtained by ex- 3.0 0.10 1.01 1.01 1.02 1.02
panding eq (13): 3.0 0.50 1.35 1.35 1.57 1.41
3.0 0.90 2.91 2.48 9.15 2.33
() a* 1 + ( 2ß2/6 2) (ß « a) (17a) 3.0 1.05 3.37 3.40 2.81
which is usually written in the form 3.0 1.65 13.65 15.30 5.48
4.0 1.05 4.59 4.65 2.81
k(T)c*1 + (ftV/32/24) (ß < a) (17b) 4.0 1.65 29.52 31.64 5.48
Bell2 discussed the three limiting forms, eq 12,14, and 15. 5.0 0.10 1.02 1.02 1.02 1.02
In the present work, by retaining the extra terms in (13) 5.0 0.50 1.47 1.47 1.57 1.41
and (16) we obtain a continuous solution that connects 5.0 0.90 3.69 3.69 9.15 2.33
these limits in a divergence-free manner. In section III we 6.0 1.05 7.26 7.35 2.81
will show that eq 13 and 16 also provide useful numerical 6.0 1.65 120.17 122.87 5.48
accuracy. 9.0 0.10 1.02 1.02 1.02 1.02
It is ¿so interesting to compare the present treatment 9.0 0.50 1.56 1.57 1.57 1.41
to the well-known result5 for an infinite (i.e., untruncated) 9.0 0.90 3.49 3.49 9.15 2.33
9.5 1.05 12.68 12.77 2.81
parabolic barrier: 9.5 1.65 1214 1217 5.48
/ a
Equations 4 5, and 6. 6 Equation 21 with V, 0. =
k(T) (18) c
,

sin / ) Equation 18: results not shown for ß > a, i. e., for
temperatures below the highest-T singularity of (18).
d
This result shows singularities whenever 1/2ß is an in- Equation 17 with fit to second derivative at top of
barrier.
tegral multiple of x, i.e., at ß = (j + 1)«, j = 0,1, 2,.... The
equations in section A show that these are a result of the the only modification needed is to replace the lower in-
artificial nature of the untruncated parabola. For example,
with =
0, the correct result (14) does not diverge. tegration limits in eq 2 and hence eq 9 by Vx instead of
zero. When this is done we obtain eq 11 but with (V0 -

Consider next the divergent terms with = 1 in eq 11. The

limiting behavior of these terms is V^ replacing V0. This gives the final result, generalizing
(13) and (16):
lim (
0—2<A
-

JUl
2a -

ß
-

exp[(/3 -

2«) V0]j )
1
=
ß 0 (19) () =

ß /a ß
exp[(/3 «)(V0 V2)] ß < a
- -

Thus there is no divergence. Furthermore, for ß = 2a, the sin {ß /a) a -

ß
=
0 term already contributes exp(/3V0/2) to k(T). For (21a)
sufficiently large a and ß, this dominates, i.e., exp(«V0)
> 2«V0, as already implied by eq 15. Although it has been ß
k(T) (exp[03
-

«)(V0 -V2)] -

1) a < ß
popular to apply eq 18 when ß « 2x [and otherwise to ß
-

a
abandon the parabolic model or to use eq 17 out of its
where
range of validity], we see that eq 13 and 16 provide a
physically and mathematically more satisfactory result that V2 =
max(V!,0) (21b)
may be used for any value of ß .
6. Extension to Unsymmetric Barriers. The derivation The definition in eq 21b allows 21a to be applied to any
in part A carries over to an exoergic barrier with no truncated parabolic barrier (i.e., Vi less than, equal to, or
modification. For an endoergic barrier greater than zero) for any « and ß.
V(x) =0 x < -(2V0/mci>2)V2 III. Applications
A. Truncated Parabola. First we wish to show that eq
V{x) =
V0- y2/n<v2x2 -{2V0/mu2)Y2 <x< 13 and 16 provide a numerically accurate alternative to
[2(V0
-

VJ/mw2]'/2 (20) numerical integration of eq 7 and 8. Table I shows a

comparison of these results for various values of the un-
V(x) =
Vi [2(V0
-

ViJ/W]1'2 < x
itless quantities V0« and ß/ . For a > ß the results are
where Vx < V0 and the particle is incident from the left, also compared to the popular eq 18. The table shows that
eq 18, which diverges at ß = a, already leads to serious
(5) See, e.g., K. A. Quickert and D. J. LeRoy, J. Chem. Phys., 52, 856 errors at ß = 0.5 for V0a < 5 and at ß = 0.9 for V0a =
(1970); W. H. Miller, ibid., 61, 1823 (1974). 9. In contrast the new analytic formula, including the
626 The Journal of Physical Chemistry, Vol. 85, No. 6, 1981 Letters

TABLE II: Transmission Coefficients for Eckart Barriers0

k(T)
uniform present un truncated Wigner
VuEr > ®-o T, K ß/^top fit, cm'1 semiclassical6 algorithm0 parabola"2 method'
-0.003 1.0e 1500 0.071 465.9 1.008 1.008 1.008 1.008
1000 0.107 465.8 1.019 1.019 1.019 1.019
600 0.178 465.8 1.05 1.05 1.05 1.05
400 0.267 465.5 1.13 1.13 1.13 1.12
300 0.356 465.4 1.24 1.24 1.24 1.21
200 0.535 464.9 1.66 1.68 1.69 1.47
150 0.713 463.6 2.71 2.80 1.84 1.71
-0.003 CO q 1500 0.214 1390 1.08 1.08 1.08 1.08
1000 0.321 1388 1.18 1.19 1.19 1.17
600 0.534 1382 1.62 1.66 1.69 1.47
400 0.802 1360 3.38 3.82 4.32 2.06
300 1.069 1233 14.24 13.89 2.88
0.009 1.0* 1500 0.063 414.5 1.007 1.007 1.007 1.007
1000 0.095 414.5 1.015 1.105 1.015 1.015
600 0.159 414.5 1.04 1.04 1.04 1.04
400 0.238 414.3 1.10 1.10 1.10 1.09
300 0.317 414.7 1.18 1.19 1.18 1.16
200 0.476 413.7 1.48 1.49 1.49 1.37
150 0.634 413.0 2.11 2.16 2.16 1.65
0.009 3.0h 1500 0.190 1236 1.06 1.06 1.06 1.06
1000 0.285 1235 1.14 1.14 1.15 1.13
600 0.476 1231 1.45 1.48 1.50 1.37
400 0.713 1218 2.47 2.69 2.81 1.83
300 0.951 1165 6.28 7.70 17.21 2.48
200 1.427 894 438.4 878.9 4.30
0.009 5.0" 1500 0.317 2043 1.16 1.18 1.15 1.14
1000 0.475 2033 1.42 1.47 1.39 1.31
600 0.792 1978 2.87 3.32 2.94 1.85
400 1.188 1633 16.15 13.70 2.92
300 1.584 1491 324.4 721.5 4.41
0
m 3474 au, V„ 0.03 Er. See footnotes 8-10 for explanation of units. b Computed by uniformized WKB method
= =

d
using program of ref 11.
c
Equations 21 and 24-26. Equation 18 with 1 : results not shown for ß > atop, i.e.,
=

for temperatures below the highest-Tsingularity of (18).

0
tvtop = 465.6 cm"1. 2 tvtop 1379 cm"1. * cvtop 412.5 cm"1. = =

h
tvtop 1237 cm"1.
=
top 1888 cm"1. ' Equation 17 with tv
1
=
tvtop.
=

terms that cancel the divergences and lead to a continuous from exp(-|8Vo) to e exp(-/3V0) when the barrier of in-
1

transition through the ß = a region, provides useful ac- terest is fit to a parabola. This yields
curacy even at ß = 1.65 . The most difficult regime for AEstr(T) =
1/( -

ß) (23)
the new formula is the case of low barrier heights; thus the
where a is computed for the fit parabola. The energy
largest errors in the table are for V0a = 1. By V0a = 3,
the new formula provides 1% accuracy up to ß = 1.05 and interval within AEstr(T) of the top of the barrier is the
by V0a = 9 it provides 0.3% accuracy up to ß = 1.65 . region that contributes most to k(T), and it is over this
B. Eckart Barrier. In this section we wish to show that region [(V0- AEstr) < E < V0] that the parabolic fit must
be best. At high T
eq 21 can also provide useful results for other, more re-
alistic, barrier shapes. Consider the Eckart potential7 AEgTR —*
1/a
T—“
Viy [Vo1/2 + (Vo Vi)1/2]2
-

The second derivative at the barrier maximum seldom

V(x) = | (22a)
leads to a good fit over even this interval, and as T de-
1 + 7 (l + y)2
creases, AEStr(T) becomes even larger.
where The general procedure we evolved to obtain a good ef-
y =
ß ( ) (22b) fective parabolic fit to the true potential at a temperature
T is as follows. Let s = w xQ where x0 is the position of
-

and 7 is range parameter. We wish to approximate this

a
the maximum in the true V(x). Consider a sequence of
by eq 20 and then to use eq 21 to approximate k(T). The distances s¡ where sx = and s,+i = s, + . For each i (i
customary procedure is to use the second derivative at the =
1, 2,...), fit a parabola (as described in the Appendix)
barrier maximum to approximate and then to use eq 17 to the true potential over the range (-s¡, +s¡). This yields
or 18 to approximate ( ). Our goal is to improve on both
co, and hence a,. We define
steps of the customary procedure.
First estimate the “significant tunneling region”
we ,· = V0- y2[V(e¡) + V(-S¡)] (24)
(AESTR, afunction of T). For this purpose it is sufficient Increment i by 1 until either of the following conditions
to consider the un-uniformized WKB approximation is satisfied:
exp(-/3E) exp[a(E V0)] to the integrand of (5). We define
-

1
AEStr as the energy interval over which this factor drops ; > (25)
di
-

min(a;,d)
(6) See, e.g., H. S. Johnston, “Gas Phase Reaction Rate Theory”, or
Ronald Press, New York, 1966. >
(7) C. Eckart, Phys. Rev., 35, 1303 (1930). 1/2mco,s¡2 min[V0,(V0 -

Vx)] (26)
Letters The Journal of Physical Chemistry, Vol. 85, No. 6, 1981 627

TABLE III: Transmission Coefficients for Fit to Scaled SSMK Potential Energy Barrier0
__

accurate uniform present un truncated Wigner

T, K ß top wfu, cm'1 quantal6 semiclassical® algorithm1* parabola® method^
1250 0.245 1303 1.14 1.09 1.10 1.11 1.10
1000 0.306 1301 1.21 1.14 1.16 1.17 1.15
600 0.510 1286 1.61 1.46 1.54 1.60 1.43
400 0.765 1245 2.79 2.42 2.83 3.57 1.96
350 0.874 1208 3.80 3.25 3.96 7.15 2.26
300 1.020 1131 6.21 5.26 6.04 2.71
250 1.224 999.0 14.74 12.45 9.17 3.46
200 1.530 854.5 83.48 71.97 19.02 4.85
0
m 1224.7333 au, b, 0.1129 eV, 6,= 0.2294 eV, 63 0.30855 a0'2, b4 1.11123 a0'2,S 0.424 eV,
= = = =
= 1337 =

cm'1. Reference 15. ® Computed by uniformized WKB method using program of ref 11. d Equations 21 and 24-26.
6
f Equation 17 with
Equation 18 with tutop: results not shown for ß > ctop.
® = =
.

The final value of , yields the effective a to use in eq 21. TABLE IV: Transmission Coefficients for
These final values are called o>ñt and am to distinguish them Gaussian Barrier0
from wjop and atop which are calculated from the second untrun-
derivative at the top of the barrier. We choose small uniform present cated
enough (0.001 Bq) so that the final results are independent w fit, semi- algo- para- Wigner
of decreasing it further. When oq > ß, eq 25 is a conse- T, K cm'1 classical rithmc bolad method®
quence of eq 23. When ß > ai: the denominator of eq 25 1250 1321 1.10 1.10
ensures that it is not used. If ß > aq, then tunneling 1000 1319 1.16 1.17 1.17 1.15
600 1312 1.52 1.57 1.60 1.43
through all portions of the barrier is important and eq 26, 3.13 3.57 1.96
which reflects the finiteness of the barrier, must be used. 400 1289 2.78
350 1262 4.05 4.75 7.15 2.26
For realistic potentials, if we did not enforce (26), as T 300 1186 7.77 8.00 2.71
decreases we would reach a point where the true potential 250 1102 27.61 21.65 3.47
at E = V0 AE¡ is concave upward. The parabolic fit is
-

200 1102 433.75 791.3 4.85

always concave downward, and attempts to fit the true 0
Barrier height = 0.0155819 Eh, m = 1224.7333 au,
potential beyond this point lead to increasingly less re- ojtop 1337 cm'1.
= 6
Computed by uniformized WKB
alistic fits. Equation 26 forces cv¡ to stop decreasing below method using program of ref 11. ® Equations 21 and
the value at which a parabolic fit is still realistic. 24-26. d Equation 18 with =
tu top: results not shown
To test this algorithm we performed calculations for a for ß > a top· e Equation 17 with =
4 .
set of Eckart barriers. We set m equal to the mass of a
deuterium atom8 and V0 equal to 0.03 Eh.9 We studied shows, as already discussed elsewhere for other cases,12-16,17
several different values for V1 and 7. The parameter that the uniformized WKB results are in reasonably good
values10 and the results are given in Table II where they agreement with accurate quantal results for the one-
are compared to accurate uniform semiclassical results and mathematical-dimensional barrier transmission problem.
to the results of the untruncated parabola formula and the A final example is provided in Table IV where we con-
Wigner method, i.e., using eq 17 or 18 with determined sider a Gaussian barrier with the same height and as
by the second derivative at the top of the barrier. The the barrier in Table III. For this shape of barrier too, the
accurate uniform semiclassical results were calculated by present procedures provide useful accuracy.
the computer program kappas11 which is described else-
where.12 We see the present method provides reasonably IV. Concluding Remarks
accurate results for a wide range of parameters. The main result of this paper is eq 21, which is a con-
As another test we considered a least-squares fit to the tinuous, divergence-free analytic expression that gives
potential energy barrier of Shavitt et al.13 (SSMK), scaled reasonable results for the Boltzmann-averaged transmis-
as suggested by Shavitt.14 This barrier is given by15 sion coefficient for a truncated parabolic barrier. We
showed that this expression may also be used to provide
V(s) = sech3 (b3s2) + b2 exp(-b4s2) +
simple analytic approximations for other shapes of barriers.
(B -

bl- b2) exp(-454s2) A popular approach in the literature is to use the parabolic
where the parameters are given in Table III. The results expression (17) to approximate the transmission coefficient
are given in Table III where they are compared to accurate for nonparabolic barriers in terms of the second derivative
quantal transmission coefficients15 and to the results ob- at the barrier maximum (which yields ). Although the
tained by applying the previous methods. Again we see conditions for validity of this formula are seldom satisfied
that the new algorithm provides considerable improvement for problems of chemical interest,18 it does often provide
over previous parabolic tunneling formulas. Table III also results of useful empirical accuracy.19 We have shown that
a much more satisfactory parabolic approximation can be
(8) 3474 atomic units (au) of mass. based on the present formula (21) and the effective par-
(9) 1 Eh = 1 au of energy = 1 hartree. Therefore, 0.03 Eh = 0.81635
eV = 18.825 kcal/mol.
(10) 1 ao = 1 au of length = 1 bohr radius = 0.52918 X 10"10 m. tie) B. C. Garrett, D. G. Truhlar, R. S. Grev, and A. W. Magnuson,
(11) B. C. Garrett and D. G. Truhlar, “National Resource for Com- J. Phys. Chem., 84,1730 (1980); B. C. Garrett, D. G. Truhlar, R. S. Grev,
putation in Chemistry Software Catalog”, Vol. 1, Lawrence Berkeley and R. B. Walker, J. Chem. Phys. 73, 235 (1980).
Laboratory, Berkeley, CA, 1980, Program No. KSOl(KAPPAS). (17) B. C. Garrett, D. G. Truhlar, and R. S. Grev, J. Phys. Chem., 84,
(12) B. C. Garrett and D. G. Truhlar, J. Phys. Chem., 83, 2921 (1979). 1749 (1980); B. C. Garrett, D. G. Truhlar, R. S. Grev, A. W. Magnuson,
(13) I. Shavitt, R. M. Stevens, F. L. Minn, and M. Karplus, J. Chem. and J. N. L. Connor, J. Chem. Phys., 73, 1721 (1980).
Phys., 48, 2700 (1968). (18) P. Pechukas in “Dynamics of Molecular Collisions”, Part B, W.
(14) I. Shavitt, J. Chem. Phys., 49, 4048 (1968). H. Miller, Ed., Plenum Press, New York, 1976, p 269.
(15) D. G. Truhlar and A. Kuppermann, J. Am. Chem. Soc., 93, 1840 (19) B. C. Garrett and D. G. Truhlar, J. Phys. Chem., 83, 200,3058(E)
(1971). (1979); 83, 1079 (1979), 84, 682(E) (1980).
628 The Journal of Physical Chemistry, Vol. 85, No. 6, 1981 Letters

TABLE A-I: Sensitivity of the Results in Table III to bolically than to calculate it fully, the expression proposed
Fitting Procedure in this paper may be very useful for practical work.
{ Acknowledgment. The authors are grateful to Dr. Bruce
uniform C. Garrett for his assistance with this project. This work
semi- 0,s;/2, was supported in part by the U.S. Department of Energy,
T, K classical6 0,±s¿/2 0,±3s,/4 ±s¡ Office of Basic Energy Sciences, through contract no.
1250 1.09 1.10 1.10 1.10 DOE-DE-AC02-79ER10425.
1000 1.14 1.17 1.16 1.16
600 1.46 1.57 1.54 1.54 Appendix 1
400 2.42 3.11 2.83 2.81 To fit a parabola to a potential over the interval (—s£,+ss)
350 3.25 4.63 3.96 3.93
6.04 6.06 where s =
0 denotes the maximum of the true potential,
300 5.26 7.38
250 12.46 31.85 9.17 9.40 we first equate a cubic polynomial to the potential at
200 71.97 1416 19.02 54.84 0,±3s£/4 (this was found by trial to give better results than
Equation 21 with fitting procedures of Appendix 1.
“ 0,±s¡/2 or 0,±s£). The cubic polynomial is written as V^ic
6
Computed by uniformized WKB method using program
=
y0 + 1/2 282 + cs3 since the first derivative is zero at
of ref. 11. s 0. The effective parabolic fit is obtained by setting
=

c = 0. The procedure just described was used for all the

abolic width of the nonparabolic barrier at the energies results in Tables II and III; however, one can sometimes
that contribute appreciably to the transmission coefficient. obtain better results by more complicated procedures. For
This allows the parabolic formulas to be used in a more example, we also tried fitting cubic polynomials to the
meaningful way. In another paper20 we shall show that potential at two sets of points 0,±s£ and 0,±s,/2 and de-
the parabolic approximation discussed here can be com- fining the best parabolic fit by averaging the quadratic
bined with a simple but realistic treatment of reaction-path coefficients obtained in the two fits. Although this five-
curvature to provide a nonempirical analytic formula for point fit sometimes gives better results, the extra com-
the transmission coefficient of atom-diatom reactions that plication does not seem warranted.
rivals in accuracy the best state-of-the-art semiclassical Table A-I shows the sensitivity of the present results to
treatments16,21 for H + H2 and isotopic analogues. Since the fitting procedure for the case shown in Table III. As
it is much easier to approximate a reaction barrier para- usual, the 0,±3s,/4 fit works better than the other three-
point fits. This case is less typical though as regards the
(20) R. T. Skodje, D. G. Truhlar, and B. C. Garrett, to be published.
five-point fit; this is the case where the five-point fit makes
(21) R. A. Marcus and . E. Coltrin, J. Chem. Phys., 67, 2609 (1977);
the most difference as compared to the 0,±3s¡/4 fit and
W. H. Miller, N. C. Handy, and J. E. Adams, ibid., 72, 99 (1980). where it offers the most improvement.