817

An equationto representgrain-sizedistribution
Murray D. Fredlund, D.G. Fredlund, and G. Ward Wilson

Abstract: The grain-size distribution is commonly used for soil classification;however, there is also potential to use
the grain-size distribution as a basis for estimating soil behaviour.For example, much emphasishas recently been
placed on the estimation of the soil-water characteristiccurve. Many methods proposed in the literature use the grain-
size distribution as a starting point to estimate the soil-water characteristiccurve. Two mathematicalforms are pre-
sentedto representgrain-size distribution curves, namely, a unimodal form and a bimodal form. The proposed equa-
tions provide methods for accuratelyrepresentinguniform, well-graded soils, and gap-gradedsoils. The five-parameter
unimodal equation provides a closer fit than previous two-parameter,log-normal equationsused to fit uniform and well-
graded soils. The unimodal equation also improves representationof the silt- and clay-sized portions of the grain-size
distribution curve.

Key words: grain-size distribution, sieve analysis,hydrometer analysis, soil classification,probability density function.

R6sum6 : La distribution granulom6triqueest utilis6e courammentpour la classificationdes sols; cependant,il est pos-
sible d'utiliser 6galementla distribution granulom6triquecomme base d'6valuation du comportementdu sol. Par
exemple, beaucoupd'emphase a 6t6 mise r6cemment sur la d6terminationde la courbe caract6ristiquesol-eau.Plusieurs
m6thodesproposdesdans la litt6rature utilisent la distribution granulom6triquecomme point de d6part pour 6tablir la
courbe caract6ristiquesol-eau.Deux formes math6matiquessont pr6sent6espour reproduire les courbes de distribution
granulom6triqe:nomm6ment, une forme unimodale et une lbrme bimodale. Les 6quationspropos6esfournissentdes
mdthodespour repr6senteravec pr6cision des sols I granulom6trieuniforme, 6tal6e et discontinue.U6quation
unimodale i cinq parambtresfournit une meilleure concordanceque les 6quationsant6rieureslog normales i deux
parambtresutilis6es pour reproduire les courbes des sols ir granulom6trieuniforme et 6tal6e.L'6quation unimodale
am6liore aussi Ia repr6sentationdes portions de silt et de grosseursargileusesde la courbe de distribution
granulom6trique.

Mots clls : distribution granulom6trique,analyse par tamisage,analyse h l'hydromEtre, classificationdes sols, fbnction
d e d e n s i t dp r o b a b i l i s t i q u e .

[Traduit par la R6dactionl

Introduction American Society for Testing and Materials standards

Dl140-54 and D422-63 (ASTM 1964a, 1964b) provide a
The grain-sizedistribution is a simple, yet informative test
basic testing and reporting method whereby the results of a
routinely performed in soil mechanicsto classify soils. Re-
sieve and hydrometer analysis are plotted on a semilog-
cent researchhas made use of the grain-sizedistribution as a
arithmic graph. An interpretation method for the series of
basis for the estimation of other soil properties such as the plotted points is specifiedin the procedure.Manual interpre-
soil-water characteristiccurve through mathematicalanaly-
tation methods,such as sketchingin a complete curve, have
sis (Gupta and Larson 1979a, 1979b;Arya and Paris l98l;
often been used to provide a completegrain-sizedistribution
Haverkamp and Parlange 1986). Mathematically represent-
curve. Gardner(1956) proposeda two-parameter,log-normal
ing the grain-size distribution provides several benefits.
distribution to provide representationof grain-size distribu-
First, the soil may be classifiedusing the best-fit parameters.
tion data. Both methodsare feasiblebut have limitations that
Second,the mathematicalequation can be used as the basis are discussedlater in the paper.
for analysisrelatedto estimatingthe soil-watercharacteristic
This paperproposestwo new models to fit grain-size data,
curve. Third, a mathematicalequation can provide a method
namely, the use of a unimodal and a bimodal mathematical
of representing the entire curve between measured data
function. The two new equationsprovide greater flexibility
points. Representingthe soil as a mathematicalfunction also
for fitting a wide variety of soils.
providesincreasedflexibility in searchingfor similar soils in
databases.
Background
Received November 20, 1998. Accepted December 3, 1999. Numerous methods have been developedfor particle-size
Published on the NRC ResearchPress website on Ausust 4. analysis in the laboratory and field. These include the
2000.
elutriation method, the test tube shaking method, the
M.D. Fredlund, D.G. Fredlund, and G.W. Wilson. Wiegner sedimentationcylinder, the photoelectric method,
Department of Civil Engineering, University of Saskatchewan, the pipette method, and the hydrometer method, in addition
Saskatoon,SK S7N 5A9, Canada. to the sieve analysis.Of these methods,only the pipette and

Can.Geotech.J.37: 817-827(2000\ @ 2000 NRC Canada

818
the soil-watercharacteristic
Table 1. Equationsthat havebeenusedto represent
Authors Equation
Gardner 1958
l4/w=l.rrg+l]ts
Brooks and Corey 1964
w.=wrt(w, -'',[f]""
van Genuchten 1980;
Burdine 1953
* (r.-'".r)l
w-=Wrb
[ 1 +t a n y1 " 1 ( ' - [ l
van Genuchten 1980;
Mualem 1976
wo=w-+(1.r.-rrr)l
van Genuchten 1980
r n ' *= w r v ! + ( w , - , " r ) [ l t
Fredlund and Xing 1994
hydrometermethodshave found generalacceptancefor fine-
grained soils (Kohnke 1968).
ASTM (1964a, 1964b) presentsa standardfor testing for
grain-size distribution. The interpretation of the grain-size
distribution is typically carried out manually. Further details
concerningthe testing procedureand the interpolationof the
sieve and hydrometer tests are provided by Lambe (1951).
Gardner (1956) used a two-parameter,log-normal distri-
bution to fit grain-sizedistribution data. Kemper and Chepil
(1965) further studiedthe work of Gardner.The two-param-
eter fit of the grain-size distribution was performed using a
geometric mean parameter,jrs, and a geometric standard de-
viation, on. The method of fitting log-normal equations to
the grain-iize distribution was not recommendedfor general
use. However, the reasongiven for not using the log-normal
method was the lack of computing power necessaryto fit the
equation to data. Hagen et al. (1987) presenteda computer-
ized, iterative procedure that required only two sieves to de-
termine the parameters for a standard, two-parameter log-
normal distribution. Unfortunately, the log-normal distribu-
C a n .G e o t e c hJ. . V o l .3 7 , 2 O O O
curve.
Definition of variables
wy saturatedgravimetric water content; Irw,
any gravimetric water content; ]rrg,
residual gravimetric water content; 4s,
fitting parameter; ns, fitting parameter; V,
soil suction
a", bubbling pressure(kPa); n., pore-size
index: w,. saturatedgravimetricwater
content:lrs. any gravimetricwater content:
w' residual volumetric water content; r{,
soil suction (kPa)
w' saturated gravimetric water content; ltw,
any gravimetric water content; wr6, residual
(, : ) gravimetric water content; a6, fitting para-
meter; r?b,fitting parameter;ty, soil suction
l , f ,tl
l w", saturated gravimetric water content; }tw,
any gravimetric water content; w-, residual
gravimetric water content; a-, fitting para-
meter; ,?m,fitting parameter; y, soil suction
l l ,* , o . *, , . , t ' -l ; l
I t w", saturatedgravimetric water content; ltw,
any gravimetric water content; w.n*, resid-
* tr*y l" y"
ual gravimetric water content; au' fitting
parameter;nu' fitting parameter;mvs,
fitting parameter;y, soil suction
rr", saturated gravimetric water content; tr;w,
any gravimetric water content; a1, fitting
parameterclosely related to the air-entry
value for the soil; nr, fitting parameter
related to the maximum slope of the curve;
[f)'']]'
{'['*.,'.
my frtting parameter related to the curvature
of the slope; /r,, parameterused to adjust
lower portion of the curve; ry, soil suction
tion often failed to provide a close fit of the grain-sizedistri-
bution at the extremesof the curve (Gardner 1956; Hagen et
al. 1987). Wagner and Ding (1994) later improved upon the
log-normal equation by presenting three- and four-parameter
log-normal equations.
Campbell (1985) presenteda classificationdiagram based
on the assumptionthat the particle-size distribution is ap-
proximately log normal. This assumptionled to the particle-
size distribution being approximatedwith a Gaussiandistri-
bution function. With this assumption,any combination of
sand, silt, and clay can be representedby a geometric (or
logarithmic) mean particle diameter and a geometric stan-
dard deviation. Values were summa"rizedin a modified U.S.
Department of Agriculture (USDA) textural classification
chart by Shirizi and Boersma (1984).
The first limitation associatedwith using a log-normal
type of equation is the assumptionthat the grain-sizedistri-
bution is symmetric. In reality, the grain-sizedistribution is
often nonsymmetric and can be better fit by a different type
of equation. Second, a method for fitting soils that are
@ 2000 NRC Canada
Fredlundet al. 819

bimodal or gap-graded is of value and the four-parameter Fig. 1. Grain-size data fit with unimodal equation for a clayey
log-normal equationshave not been found to be satisfactory silt: (a) best-fit curve, R2 = 0.998; (b) arithmetic probability den-
for fitting these types of grain-size distribution. sity function; (c) logarithmic probability density function (soil
There are three general categories of grain-size distribu- number 10030).
tions (Holtz and Kovacs 1981): well-graded soils, uniform (a) too
soils, and gap-graded soils. This paper focuses on these
three categoriesof grain-sizedistribution and provides equa- 80
tions to fit the experimental data for each category. The s
well-graded and uniform soils are examined using a o o
.E
u

unimodal method of fitting an equation, and then a mathe- 6

matical meansof representinga gap-gradedsoil is presented. o 4 n .

o
f?
3 2 0
Unimodalequation for grain-size
distribution data 0 ]*
1
0.0001 0.001 0.01 0.1
The selectionof an appropriate,mathematicalequationin- Particlesize (mm)
volved a review of a variety of equationsthat could be used
*M.Fredlund Unimodal - - USCS o/oClay --- USCS % Sand
to frt soils data. It has been observed that the soil-water char-
acteristic curve possessesa shape similar to that of the r Experimental -'- USCS % Silt
grain-size distribution curve. This is probably to be ex-
pected, since the soil-water characteristiccurve provides a
(b) oooo
representationof the void distribution in a soil, whereasthe
grain-size curve provides information on the distribution of 5000
the solid phase of the soil. Since the solids plus the voids
add up to the total soil volume, it is to be expectedthat the 4000
E
distribution of the solids phase (i.e., grain-size distribution) €
would tend to bear an inverserelationshipto the distribution 3000
:
of voids (i.e., representedby the soil-water characteristic E
-
2000
curve), and vice versa.
A summary of several of the equations that have been 1000
used to fit the soil-water characteristiccurve is given in Ta-
ble l. Brooks and Corey (1964) and Gardner (1974) pte- 0
0.0001 0.001 0.01 0.1 1 10 100
sentedthree-parameterequationsand van Genuchten(1980)
Particle size (mm)
and Fredlund and Xing (1994) presented four-parameter
equations.It would seem reasonablethat a form of equation -*- Particle-sizeArithmetic -. - USCS % Silt . Particle diameter
similar to those shown in Table I could be used to represent -'-USCS%ClaY ---USCS%Sand

the grain-size distribution.

An accurate representation of the clay fraction of the
grain-sizedistribution was considerednecessaryto complete (c) too
the mathematical function. Since the Fredlund and Xing
(1994) equation allows independentcontrol over the lower 80
end of the curve (i.e., the fine particle size range), it was se- E
o A n
lected as the basis for the developmentof a grain-sizedistri- ' ca v v

bution equation. The reversed scale of the grain-size o

o ,^
distribution and characteristicsunique to the grain-size dis- z +
o
u

tribution requiredthe original Fredlund and Xing equationto co

o- 20
be modifred to the form shown as follows:

0+-
tll Pe(A= 0.0001

i,f'"['.g)l' -----. Particle-sizeLog PDF

Particlesize (mm)

r Experimental - - USCS % Sitt

l L;F*ll
' ' ' ' . M . F r e d l u n dU n i m o d a l - -USCS % Clay - - - - - " -U S C S % S a n d

t t (- \'..11-"
j t n l e x .n t t r l. i[ ? l
t L ll an, is a parameterdesignatingthe inflection point on the
curie and is related to the initial breaking point on the
wnere curve:
Po(d) is the percentage,by mass, of particles passing rzn,is a parameter related to the steepestslope on the
partrcularslze; curie (i.e., uniformity of the particle-sizedistribution);
@ 2000 NRC Canada
820 C a n .G e o t e c hJ, . V o l .3 7 , 2 O O O

mn is a parameter related to the shape of the curve as it The particle-size distributions presented in this paper are
appr-oaches the fines region: calculated using eq. 12) and are referred to as the arithmetic
d.*, is a parameterrelated to the amount of fines in a soil; probability density function. Figure 1b illustrates the arith-
d is the diameter of any particle size under consideration; metic probability density function for the clayey silt (soil
and number 10030) shown in Fig. la.
d* is the diameterof the minimum allowable size particle. The highest point in the PDF plot is the mode or the most
Equation [1] is refened to as a unimodal equationand can frequent particle size. Since eq. l2l is a PDF, the natural
be used to fit a wide variety of soils. A quasi-Newtonfitting laws of probability hold, such that the area under the differ-
algorithm was used to adjust three of the four parametersto entiatedcurve must equal unity:
fit the equation to each soil. The algorithm progressively
minimizes the squareddifferencesbetweenthe equation and *.-
/ dP^
n )l d r = l
experimental data. The best-fit particle-size distribution f3l ll
function can be plotted along with the grain-sizedistribution :_\dd )
data, typically on a logarithmic scale, as shown in Fig. la
for a clayey silt (soil number 10030).t Equation l2l can be arithmetically integrated between the
The unimodal equationprovides significant improvements specifiedparticle-diametersizes. The probability that a soil
in the fit of grain-size data over previous mathematical particle diameter will fall in a certain range is determined by
representations(i.e., log-normal distribution). This is to be the following relationship:
expected due to the increase in the number of parameters
used to representthe grain-sizedistribution. The complexity r=d z
of the proposed unimodal equation due to the added parame- probability (d( d < dz) =
l4l | p@)d,
ters is determined to be insignificant becauseof the avail-
x=d t
ability of curve-fitting software.
The particle-sizedistribution provides information on the
amount and dominant sizes of particles present in a soil. It is convenient to represent the PDF in a different manner
However,anotherform can also be used to representthe dis- when plotted on a logarithmic scale. The arithmetic PDF
tribution of particle sizes by differentiating the particle-size will often appear distorted when plotted on a logarithmic
distribution curve. The differentiation produces a particle- scale.The peak computedfrom eq. [4] will not representthe
size probability density function (PDF). The differentiated most frequent particle size. To overcomethis limitation, the
form of the unimodal grain-sizeequation is given in eq. [2], PDF is often representedby taking the logarithm of the par-
and the parameterspresentedin the particle-size PDF are the ticle size and differentiating the grain-size equation to pro-
same as those defined for eq. []: duce a PDF which appears more physically realistic as
presentedin eq. [5]:
rll
dP^
P -
t L l
dd dP^ dP^
Pln(10)d
l5l p,(d)= =
d log(d dd

rf r* 4-'l I'l

Jf'"[*)"
;t'-el d ) where pr(d) is the logarithmic PDF.
\
The peak of eq. [5] will representthe most frequent parti-
I l- , ',",11'"' cle size. It must be noted that the probability of the logarith-
mic PDF must be calculatedaccording to eq. [6]:
J r n l e x\ "u'/|r .llf?
t L .ll x=tog @ z)

t6l (d( d < dz) =

probability | pt@)a-r
r=tog (dr )

. [+). *[*n,',
. [+). Figure lc shows the logarithmic PDF for the clayey silt
(soil number 10030).
[*ou, ] ] The unimodal equation fit for a silty sand (soil number
63) and a sandy clay (soil number 11648) are shown in
Figs. 2a and 3a, respectively. Also shown are the arithmetic
,"fr*4-.]"
d ) d,r,
probability density functions (i.e., Figs. 2b,3b) and the loga-
\ rithmic probability density function for each of the above
t t , .n.-ll'"'
I t n l e x p\ rdt r)| .jll+lJ
''['.*)'[,'['.9)] s o i l s ( i . e . .F i g s .2 c . 3 c ) .
The variation of R2 (where R is the corelation coefficient)
versuspercent clay is shown in Fig. 4. The value of R2 was
I L plotted versus percent clay because the representationof
fines by the new equation was consideredimportant.

I Soil numbers refer to soils found in the SoilVision database,which is a proprietary product of SoilVision Systems Ltd.

O 2000NRC Canada
Fredlundet al. 821

Fig. 2. Grain-size data fit with unimodal equation for a silty Fig. 3. Grain-size data fit with unimodal equation for a sandy
sand: (a) best-fit curve, R2 = 0.985; 1b; arithmetic probability clay: (a) best-fit curve, R2 = 0.999; (b) arithmetic probability
density function; (c) logarithmic probability density function (soil density function; (c) logarithmic probability density function
number63). (soil number 11648).

(u)tooi (a) 1oo

BO 80
s s
' c6 60 ' c6
a
o a
G
o 40
E 40
c) c
o
I
o
20 o
(L

;
0; 0
0.0001 0 . 0 0 1 0.01 '1 10 100
0.0001 0 . 0 0 1 0 . 0 1 0.1 0.1
Particle size (mm) Particlesize(mm)
- M . F r e d l u n d U n i m o d a l - - U S C S o / oC l a y - - -
USCS % Sand *M.FredlundUnimodal- . - USCS% Clay ---
USGS% Sand
r Experimental - - USCS%Silt
r ExDerimental - -'USCS % Silt

(b; sooo
(b; +oo 4500
350 4000+
300 3500+
250 3000
E E
E c 2500
200
2000+
150 1500
100 1000
50 500
0
0.0001 0.001 0.01 0.1 1 10
0.0001 0.001 0.01 0.1 1 10
Particlesize(mm) (mm)
Particle-size
- Particle-size
ArithmeticPDF - Particle-size
ArithmeticPDF _ . _ .USCS % Silt
_ . _ .USCS % Silt
- ..USCS%Clay --'USCS%Sand -..-USCS%Clay -- USCS%Sand

(c; too (c) 1oo

;<
80
;c s
o o u o o u
@
G .E
o
= - a n
o -
8 + o
co zo
P
o 2 0
8 2 0

0 '1
0.0001 0.001 0.01 0.1 10 100
0.0001 0.001 0.01 0.1 1
Particlesize(mm)
Particlesize(mm)
------.Particle-size
LogPDF r Experimental - . - .USCS% Sitt
--..-.. Particle-sizeLog PDF r Experimental -.-.uscs%silt '''" [/.FredlundUnimodal - - USCS% Clay -'-"' USCS% Sand
' ' " " [ , 4 . F r e d l u n d U n i m o d a- l. ' - U S C S % C l a y ------- USCS % Sand

Parametricstudy of the proposed unimodal curve. The parameterc' is related to the initial break of the
grain-size distribution equation equation and is more precisely the inflection point on the
curve. Its effect on the grain-size distribution curve can be
A parametric study of the proposed unimodal equation seenin Fig.5a, where dn. is varied from 0. I to 10 while the
(i.e., eq. []) showsbehaviour similar to that of the Fredlund other equation parameteisare held constant.The parameter
and Xing (1994) equation for the soil-water characteristic ar. provides an indication of the largest particle sizes.
@ 2000 NRC Canada
822 C a n . G e o t e c h .J . V o l . 3 7 , 2 O O O

Figure 5b shows how the parametern* influences the slope Fig. 4. Variation of R2 enor as the amount of fines represented
of the grain-size distribution. The point of maximum slope in a soil increases.
along the grain-sizedistribution provides an indication of the 1
dominant particle size (i.e., on a logarithmic scale) in the
0.99
!F6mF;mn.%
'.xw
o o o
soil. In the parametricrepresentation,n' is varied from I to 4. a' o
0.98
The parameterzr. influences the break onto the finer par- ooe o
o
ticle size of the sample. The effect of varying the parame- 0.97 o
5 o^o
term%from 0.3 to 0.9 can be seenin Fig.5c. The parameter N
E
^ ^ ^
u.Yo , a
o t
d.n, alfects the shape along the finer particle size portion of 0.95
the curve. However,the influence on the curve is quite mini- 8'
0.94
mal as shown in Fig. 5d. In some casesdrn,can be modified
to improve the fit of the overall equation. With the best-fit 0.93
o
analysisshown,drn.was adjustedmanually to improve the fit 0.92
of the curve to the data. It was found that a value of 0.001 20 40 60 80 100
for dr* provided a reasonablefit in most cases. %ClaY

Bimodal equation for the grain-size

distribution curve
There is a limitation in using the unimodal equation (i.e.,
eq. [1]) when the soils are gap-gradedas shown in Fig. 6. In
this case, it is necessaryto consider the use of a bimodal
equation when performing the best-fit analysis. Soils fre-
quently have particle-sizedistributions that are not consis-
tent with a unimodal distribution and, as a result, attemptsto
fit the unimodal equation to certain data sets can often lead
to a misrepresentationof the character of the particle-size
distribution.This is particularly important when the equation
is differentiatedand used for further analyses(e.g., estima-
tion of the soil-water characteristiccurve).
The characteristicshapeof a bimodal or gap-gradedsoil is
"hump" often observed from experimental data.
the double
These humps indicate that the particle-size distribution is
concentrated around two separate particle sizes. From a
mathematicalstandpoint,a gap-gradedsoil can be viewed as
a combination of two or more separatesoils (Durner 1994).
"stackins" of more than one unimodal
This allows for the
equation:

Fig. 5. Parametervariation: (a) effect of varying the parametera*. while nr,= 4.0,mr,= 0.5, d,er = 1000, andd^= 0.001; (b) effect of
varying the parameternr. while a*. = 1.0,mr, = 0.5, d,g. = 1000, and d. = 0.001; (c) effect of varying the parameterm* while ar, =
1.0,r?sr= 4.0, dry, = 1000, andd, - 0.001; (d) effect of varying the parameterd.r, while a*. = 1.0,nr,= 4.0,mr,=0.5, andd- = 0.001.
100 100
(a) (b)
80 80
E ;<
;
.B oo q
@
6 6
o o
E
o
a ,- n 6 + u
o 6
(!
20 20

0
0.1 1
Particlesize (mm)

'100 100
(c) (d)
80 80
t E
a - - f 6 0
6 o
= A i
E a o b - "
o -
P Y
o
(L
o
(L
20 20

O 2000NRC Canada
Fredlundet al. 823

I7l Pp(A= Fig. 6. Example of fit of a gap-gradedsoil with the unimodal

equation (R2 = 0.977) and the bimodal equation (R2 = 0.999)

k j,lg-rll'1 (soil number ll49l).

I L'{'.*]l
Lwi
t t , ',-ll'''
]rnl.^n,r.l9l
L L ' )lf) l E
P
'6

6
o
o-
E
6 0

40
o)
3 z o
(L
where
d. is the residual particle diameter;
"subsystems"for the total particle-size 0.01 0.1 I 10 100
k is the number of 0.001
distribution; and Patriclesize(mm)
w; are weighting factors for the subcurves, subject to 0 <
wr<land2w,=1.
' - - - - -M . F r e d l u n dB i m o d a l r Exoerimentap- -USCS % Silt
" " " M . F r e d l u n dU n i m o d a l - . - U S C S ' / o C l a v - - - U S C S % S a n d
For a bimodal curve, k would be equal to 2 and the num-
ber of parametersto be determinedis 4k + (k - 1) (i.e., 9).
The unimodal equationis used as the basis for the prediction A total of nine parametersmust be computedwhen fitting
of the bimodal equation. The final equation for a bimodal the bimodal equation to data (i.e., stacking).Sevenparame-
curve is siven as follows in its extendedform: ters can be determinedusing a nonlinear least squaresfitting
algorithm, and two parameterscan be essentiallyfixed (i.e.,
dr6i and d-).
Bimodal data sets can be closely fit using the bimodal
equation best-fit analysis. However, the best fit using a
f8l P^(d)= unimodal equationprovided only an adequatefit of the same
| | t- \'"ll^ data sets.On the other hand, if the data sets are unimodal in
l r n l e x p\r a
r rt* l +
l fl character,it is better to fit the data using the unimodal equa-
t L lJ tion. The superpositionmethod provides a robust method of
fitting bimodal data sets. The results of fitting the bimodal
curve to several different soils can be seen in Figs.6-10.
The R2 values for all bimodal fits was 0.999.
Figure 6 shows a comparisonof a bimodal and unimodal
+ (l -w) best fit of a data set. The unimodal R' value is 0.977 as op-
posed to 0.999 for the bimodal best fit. Comparablereduc-

{'"1*".(+)']}'
( -
iions in the R2 value are shown for the soils in Figs. 8-10.
The bimodal silty sand shown in Fig. 7 (i.e., soil number
11492) also illustrates the bimodal particle sizes on the
arithmetic and logarithmic probability density function. The
I r li
analysis shows that it is the logarithmic probability density
I l"['.4")
d 1I l function that provides the most meaningful representationof
" L - l\ the dominant particle size. In this casethe dominant particle
sizes are approximately0.008 and 0.5 mm.
I lilr*4.,-ll
d^))
I L\ Application of the mathematical function
for the grain-size distribution
where
a61 is a parameter related to the initial breaking point The grain-sizedistribution has been used primarily for the
along the curve; classificationof soils. The use of a mathematicalequationto
noi is a parameter related to the steepest slope along the fit the grain-sizedistribution provides severaladvantagesfor
curve; geotechnicalengineering.First, the unimodal and bimodal
m,oiis a parameterrelated to the shapeof the curve; equationsproposedin this paper provide a method for esti-
j6i is a parameter related to the second breaking point of mating a continuousfunction. Second,soils can be identified
the curve; on the basis of grain-size distribution by equationsthat are
k6i is a parameterrelated to the secondsteep slope along best fit to the data. This information can be stored in a data-
the curve; base and used for identification purposes.Third, equations
161is a parameter related to the second shape along the provide a consistentmethod for determiningphysical indices
curve; and such as percent clay, percent sand,percent silt, and particle-
d.6;is a parameterrelatedto the amount of fines in a soil. diameter variablessuch as drc, dzo,dzo,dso, and d6s'

@ 2000 NRC Canada

824 C a n . G e o t e c h .J . V o l . 3 7 , 2 O O O

Fig. 7. Example of fit of a gap-gradedsoil with the bimodal
equation for a clayey, silty sand: (a) best-fit curve, R2 = Q.999;
(b) arithmetic probability density function; (c) logarithmic proba-
bility density fu nction (soil number 1l 492) .
soil nt
Fig. 8. Example of fit of a gap-gradedclayey, silty sandy soil
with the unimodal equation iR2 = O.eS+)and the bimodal equa-
tion (R2 = 0.999) (soil number 11492).

(a) too -E--1- J

t!
!l E'- t s
en
BO . : -- ,r-,,-'|,1-
r i,,-..1.,
i 1 '6
E i : ^--^ I )i
/ l
l a
bu
--^t--------*f
r--
(E
'6 ;' : i t
{ i i E 4 0
a
G t r) o
I -r o
-o 40
c
o i i /, 4*{
f i
0)
o 2 0

r ft ' '
h ^ ^
. ,
d z w
ii
r l : 0.0001 0.001 0.01 0.1 1 10 10
'10
0.0001 0 . 0 0 1 0.01 0.1 1 100
Particlesize(mm)
Particlesize(mm) "--"--M.Fredlund
Bimodal r Experimental- -USCS % Silt
-M.Fredlund Unimodal -. - USCS % Clay --- """ M.Fredlund
USCS % Sand Unimodal- -USCS% Clav --- USCS% Sand
. Experimental - - uscs%sirr

Fig. 9. Example of fit of a gap-gradedsoil with the unimodal

(b;+ooo equation (R2 = 0.992) and the bimodal equation (R2 = 0.999)
3500 (soil number 11493).

3000
2500
E ': 80
E ;<
2000
p
1500 '6
bu
o
o
o
d "
P
o
L 2 0
0.0001 0.001 0.01 0.1 1
Particlesize(mm)
0
-'*" Particle-size
ArithmeticPDF * . - .uscs o/ositt 0.0001 0 . 0 0 1 0.01 0.1 1
-...USCS%Clay - - USCSo/oSand
Particlesize(mm)

Bimodal r Experimental- *USCS o/oSilt

--'--'M.Fredlund
(c) too frrtr
""" M.Fredlund Unimodal- . -USCS ohClav --- USCS% Sand

80
s
' c6 60 Fig. 10. Example of fit of a gap-gradedsoil with the unimodal
o
o equation (R2 = 0.987) and the bimodal equation (R2 = 0.999)
o
c 40 lsoil number 11498).
0)
d
100
20
80
0 E
0.0001 0.001 0.01 0.1 1 10 100
.g
o
60
Particlesize(mm) o
(6
o
"-"-... Particle-sizeLog PDF r Experimental - . - .USCS % Sitt E
€) 4 0
'''''''''' M.FredlundUnimodal - - USCS % Clav ------- USCS % Sand
S 2 0

The grain-sizedistribution has also been shown to be cen- 0

tral to several methods of estimating the soil-water charac- 0.0001 0 . 0 0 1 0 . 0 1 0.1
teristic curve (Gupta and Larson 1979a, 1979b; Arya and
Particlesize (mm)
Paris 1981; Haverkampand Parlange1986; Ranjitkar 1989). "."-"M.Fredlund
An accuraterepresentationof the soil particle sizesis essen-
Bimodal r E x p e r i m e n t a*l - U S C S % S i l t
""" M.Fredlund Unimodal ' -USCS o/oClay --'USCS % Sand
-
tial when the srain-size distribution curve is used as the

O 2000 NRC Canada

 Fredlundet al. 825

Fig. 11. Unimodal parameter variation: (a) frequency distribution Fig. 12. Determination of the soil fractions (i.e., percent clay,
of the natural logarithm of the parameter ar,; (.b) frequency dis- silt, and sand), according to the USDA classification,when using
tribution of the parameternn.; (c) frequency distribution of the the unimodal equation.
parametermsr. 100
18
,'
(a)
16
E I ;oarse
14 I
.P 60
o
I Sand
.=
;<
ta €
q I
5'o
c
c r v
o
co f
qJ ,
o
E 20 r_ sitt
L!
o Clat
0
4
0 . 0 0 0 0 1 0.0001 0.001 0.01 0.1 10 100
size(mm)
Particle

Parametersof the grain-size distribution

equations
(b) The unimodal fit of the grain-sizedistribution has been fit
30 to over 600 experimentally measured grain-size data sets
containedin the SoilVision@database.The unimodal fit per-
z3 formed well with the exception of soils exhibiting bimodal
s behaviour. The parameters of the unimodal equation appear
> 1 V to vary in a manner similar to that of the parameters in the
o
c
o Fredlund and Xing (1994) soil-water characteristic curve
* r < equation.
6 ' "
!. t! Histograms showing the frequency distribution of
10 unimodal equationparametersare shown in Fig. 11. The fre-
quency distributionsprovide an indication of the mode and
range of each of the three main parametersfor the soils con-
tained in the database.For example, the most frequent n*
and mn values were approximately 3 and 1, respectively.
0.0 2.0 4.0 6.0 8.0 1 0 . 0 1 2 . 0 1 4 . 0 1 6 . 0 1 80
One aspect of this study was to determine whether the
log (rb,) equation parameterscould be grouped accordingto soil tex-
tural classifications.For example,is there a range of the pa-
35 rameter no, typical for silty sands? The results of this
(c) researchindicate that generalparametergroups can be iden-
30
tified but specific parametergroupings cannot be identified.
The influence of equationparameterson each other does not
ZJ
allow for specific groupings.For example,the parameternr.
s influences the parametermn, and vice versa. It was found
c that grouping soils was more successfulwhen parameters
o
* r c with physical significancewere selected.The groupings of
6 ' "
tL
soil propertieswas better achievedby grouping soils accord-
10 ing to physical parameterssuch as percentclay, percent silt,
and percentsand,or using variablessuch as dt0, d20,dzo,dso,
and d6s.

Determining physical parameters from the

1.0 1.5 2.0 2.5 3.0 3.5 4.0
grain-size distribution equation
tog (nb,)
One of the benefits of the two grain-size equationspre-
sented in this paper is that meaningful, physical variables
basis for the estimation of the soil-water characteristic can be computedfrom the curves.The most commonly used
curve. The equationspresentedin this paper appearto pro- variables are percent clay, percent sand, percent silt, and di-
vide an excellent basis for the estimation of the soil-water ametervariablessuch as dto, dzo,dzo,dso, and d6s.The equa-
characteristic curve (Fredlund et al. 1997). tions are in the form of percent passing a particular particle

O 2000 NRC Canada

826
Fig. 13. Determination of the percent passing for any particle
size, d, for a unimodal grain-size distribution.
100
,'
80
- u60
t of magnitude.
4 6 0 I
.g oqn A Gonclusions
o
,
o
C
4 0
o
6 -.J'
t 2 0
A
20
u10
0 +--
0.00001 0 . 0 0 0 1 0 . 0 0 1 0.01 0.1
Particlesize(mm)
size, Po(d), where d is the particle diameter (mm). The per-
cent cl^ay,percent silt, and percent sand can therefore be
read directly from the curve by substituting in the appropri-
ate diameters.The diametersused depend upon the criteria
associatedwith the various classification methods. For ex-
ample, the USDA classificationboundariesare 0.002, 0.05,
and 2.0 mm for percent clay, percent silt, and percent sand,
respectively.The Unified Soil ClassificationSystem(USCS)
classificationusesboundariesof 0.005, 0.075, and 4.75 mm
for percentclay, percent silt, and percent sand,respectively.
The divisions can be determined for any classification
method by substitutingthe appropriateparticle size into the
equations,as shown in Fig. 12.
The diametervariablesmust be read off the curve in an in-
verse manner. The particle-size diameter answers to the
question "what particle diameter has l07o of the total mass
smaller than this size?" Taking the inverse of either the
"half-length"
unimodal or bimodal equation is difficult. A
algorithm was therefore used to read diameters from the
grain-sizecurve. An initial estimationdiameterwas selected
and the correction distance was progressivelyhalved until
the iteration processyielded a minimal error. The results of
this processare shown in Fig. 13.
Application to the estimation of the soil-
water characteristic curve
Several current methods for the estimation of the soil-
water characteristic curve make use of the grain-size distri-
bution as the basis for the prediction (Arya and Paris 1981;
Haverkamp and Parlange 1986). The soil-water characteris-
tic curve has been found to be sensitive to the derivative of
the grain-sizedistribution (Fredlundet aL.1997).Methods of
estimating the soil-water characteristic curve have been
somewhatlimited by the lack of a continuousgrain-sizedis-
tribution function. Current models often require grain-size
information to be estimated from the grain-size curve.
The grain-size distribution is theoretically a continuous
curve representingthe amount of various particle sizespres-
ent in soil. The soil-water characteristiccurve is primarily a
representationof the pore sizespresentin the soil. A method
for translating the particle-sizedistribution into a pore-size
distribution is central to most methodsof estimatingthe soil-
water characteristiccurve. An analysis of the soil can be
C a n .G e o t e c hJ. . V o l .3 7 , 2 O O O
performed by differentiating the grain-size equation as
shown by eq. [2]. The differentiatedequationwill produce a
particle-sizeprobability density function that can be used as
the basis for further analysis.The particle-sizedistributions
calculatedaccording to eq. [1] can vary over severalorders
Fitting of the grain-size distribution has historically been
a manual processor has involved the use of log-normal dis-
tributions of one, two, or three parameters.Unimodal and bi-
modal equationsare presentedin this paper to fit essentially
any grain-size distribution data set. The unimodal equation
1 100 was found to provide a good fit for a variety of soils. The
extremesof the grain-sizedistribution were also well fit by
the equation.
Gap-gradedsoils can be best fit using a bimodal equation.
The bimodal equation allows for a mathematical representa-
tion of any grain-sizedistribution where the samplecontains
two distinctly different, but dominant particle-sizegroups.
Mathematical representationof the grain-sizedistribution
provides numerousbenefits. First, the grain-size curves can
be identified and categorized.Likewise, the grain-sizecurves
can be located in a databaseusing searching techniques.
Grain-size variables (i.e., percent clay, drc, d6g,etc.) can be
mathematically determined from the equation. The unimodal
and bimodal equations provide a method for fitting the three
major soil categoriesof well-graded soils, uniform soils, and
gap-gradedsoils.
The proposed continuous mathematical function for the
grain-size curve sets the stage for further analysis to esti-
mate the soil-water characteristiccurve of a soil.
Acknowledgements
The authors would like to thank Sai Vanapalli and Doug
Charleson for their assistancewith the literature review. The
authors would also like to acknowledge the assistanceof
Mrs. Noshin Zakerzadeh in the preparation of this paper.
References
Arya, L.M., and Paris,J.F. 1981. A physicoempiricalmodel to pre-
dict the soil moisture characteristicfrom particle-size distribu-
tion and bulk density data. Soil Science Society of America
Journal,45: 1023-1030.
ASTM. 1964a. Standardmethod of test for amount of material in
soils finer than the No. 200 sieve (D1140-54) (adopted 1954).In
Nomenclature and definitions standard methods suggestedmeth-
ods. 4th ed. American Society for Testing and Materials, Phila-
delphia, Pa., pp. 18-79.
ASTM. 1964b. Standard method for grain-size analysis of soil
(D422-63) (adopted 1963). In Nomenclature and definitions
standardmethods suggestedmethods.4th ed. American Society
for Testing and Materials, Philadelphia,Pa., pp. 95-106.
Brooks, R.H., and Corey, A.T. 1964. Hydraulic properties of po-
rous media. Colorado State University, Hydrology Paper 3 (21).
Burdine, N.T. 1953. Relative permeability calculations from pore
size distribution data. Journal of Petroleum Technology, 5(3):
7 l-78.
Campbell, G.S. 1985. Soil physics with basic.Elsevier,New York.
@ 2000NRC Canada
Fredlundet al.
Dumer,W. 1994.Hydraulic conductivityestimationfor soils with
pore structure.WaterResourcesResearch,30(2):
heterogeneous
2rr-223.
Fredlund,D.G., and Xing, A. 1994.Equationsfor the soil-water
characteristiccurve. CanadianGeotechnicalJournal,3ll. 521-
532.
Fredlund,M.D., Fredlund,D.G., and Wilson,G.W. 1997.Predic-
tion of the soil-watercharacteristiccurve from grain-sizedistri-
bution and volume-massproperties./z Proceedingsof the 3rd
BrazilianSymposiumon UnsaturatedSoils,Rio de Janeiro,22-
25 April 1997,Vol. I, pp. 13-23.
Gardner,W.R. 1956.Representation of soil aggregatesizedistribu-
tion by a logarithmic-normaldistribution.Soil ScienceSociety
of AmericaProceedings, 20: 151-153.
Gardner,W.R. 1958.Somesteadystatesolutionsof the unsaturated
moistureflow equationwith applicationto evaporationfrom a
watertable.Soil Science,ESg):228-232.
Gardner,W.R. 1974.The permeabilityproblem.Soil Science,117:
243-249.
Gupta,S.C.,and Larson,W.E. 1979a.Estimatingsoil-waterreten-
from particlesize distribution,organicmatter
tion characteristics
percent,and bulk density. Water ResourcesResearch,15(6):
1633-1635.
Gupta,S.C.,andLarson,W.E. 1979b.A modelfor predictingpack-
ing densityof soils using particle-sizedistribution.Soil Science
Societyof AmericaJournal,43:758-764.
Hagen,L.J., Skidmore,E.L., and Fryrear,D.W. 1987.Using two
sievesto characterizedry soil aggregatesize distribution.Trans-
actions of the American Society of Agricultural Engineers,
30(1):162-165.
827
Haverkamp,R., and Parlange,J.Y. 1986. Predicting the water-
retentioncurve from a particle-sizedistribution:1. Sandysoils
without organicmaner.Soil Science,142(6):325-339.
Holtz, R.D., and Kovacs, W.D. 1981. An introduction to
geotechnicalengineering.kentice-Hall, Inc., EnglewoodCliffs,
N.J.
Kemper,W.D., and Chepil, W.S. 1965.Size distributionof aggre-
gates./n Methodsof soil analysis.Part l. Editedby C.A. Black.
Agronomy,9:499-510.
Kohnke,H. 1968.Soil physics.McGraw-Hill Book Company,New
York, N.Y.
Lambe,W.T. 1951.Soil testing.John Wiley & Sons Inc., New
York. N.Y.
Mualem, Y. 1976.A new model for predictingthe hydrauliccon-
ductivity of unsaturatedporous media. Water ResourcesRe-
search,12:513-522.
Ranjitkar,S.S.B. 1989.Predictionof hydraulicpropertiesof unsat-
uratedgranularsoilsbasedon grain sizedata.Ph.D. thesis,Uni-
versity of Massachusetts.
Shirizi, M.A., and Boersma,L. 1984.A unifying quantitativeanal-
ysis of soil texture.Soil ScienceSocietyof AmericaJournal,48:
142-1.47.
van Genuchten,M.T. 1980.A closedform equationfor predicting
the hydraulicconductivityof unsaturated soils. Soil ScienceSo-
ciety of America Journal,44: 892-890.
Wagner,L.E., and Ding, D. 1994.Representing ag$egatesizedis-
tributions as modified log normal distributions.Transactionsof
the American Society of Agricultural Engineers,37(3): 815-
821.
@ 2000 NRC Canada