A UNIVERSAL SCALING LAW

FOR NANOINDENTATION, BUT NOT ONLY

Nicola M. Pugno
Department of Structural Engineering, Politecnico di Torino,
Corso Duca degli Abruzzi 24, Torino, 10129, Italy

In this letter we derive a universal law for nanoindentation, considering different sizes and shapes of
the indenter. The law matches as limit cases all the well-known hardness scaling laws proposed in the
literature. But our finding can also explain their deviations experimentally observed at the nanoscale.
An even more general scaling law is then formulated, also in the fast and slow dynamics; it is based
only on the surface over volume ratio of the domain in which the energy flux occurs: thus, its
application in different fields, also for chaotic and complex (e.g., biological) systems, is demonstrated.

PACS number: 62.20.-x; 61.72.Lk; 62.20.Fe; 62.20.Qp

Hardness is defined as the load divided by the projected area of the indentation, thus it is the mean
pressure that a material will support under load. This parameter is only “nominally” a constant and is
experimentally dependent on penetration depth, size and shape of the indenter. A variation in hardness
versus penetration depth is usually defined, and perhaps not properly, as indentation size-effect,
whereas we should refer to the variation of hardness by varying the size or shape of the indenter as true
size- or shape-effects respectively. Much of the early work on indentation was reviewed by Mott
(1956). Ashby (1970) proposed that geometrically necessary dislocations (Nye, 1953) would lead to an
increasing in hardness measured by a flat punch. The problem to a conical indenter has been
investigated by Nix and Gao (1998), showing a good agreement with microindentation experiments.
However, recent results that cover a greater range of depths show only partial (Poole et al., 1996;
Swadener et al., 2002) or no agreement (Lim and Chaudhri, 1999) with their model. Thus, such a
model was extended by Swadener et al. (2002) in a very interesting way, to treat indenters of different
sizes and shapes; the results were compared with microindentation experiments, but limitations for
small depths of pyramidal indenters and sizes of spherical indenters were observed at the smaller size-
scales. The aim of this letter is the developing of a new model capable of matching as limit cases the
discussed indentation laws, simultaneously capturing the deviation observed at the nanoscale.
Incidentally, an extension of the Taylor’s law (1938) is formulated. Moreover, a very general scaling
law is provided, that seems to promise interesting applications in different fields, such as engineering,
physics, biology, medicine, economy, to cite a few, also for complex and chaotic systems and also in
the fast and slow dynamics.
Consider an indenter with a given geometry h = h(r , ϑ ) , with r and ϑ polar coordinates. The Nix
and Gao (1998) and Swadener et al. (2002) models assume that plastic deformation of the surface is
accompanied by the generation of geometrically necessary dislocation loops, in our treatment of length
l (h ) , below the surface; the deformation volume is assumed to be an hemispherical zone below the
contact area A with radius a = A π or volume V = 2π 3 ( A π ) (Figure 1). Thus, the total length
32

L of the geometrically necessary dislocation loops can be evaluated by integrating the number of steps
on the indented surface (see Figure 1):
 h
l (h ) dh =
1 S
L=
b0∫ b
(1)

where S is the lateral surface area of the indented zone, or of the indenter itself for monotonic positive
defined variation h vs. r (Figure 1) and b is the (modulus of the) Burger’s vector. Note the generality of
the result in eq. (1), that do not need the specification of the form of h, as required by the previous
models. Thus, the average geometrically necessary dislocation density is:

L S
ρG = = (2)
V bV

Arsenlis and Parks (1999) have shown that the actual number of dislocations that must be generated to
accommodate plastic deformation is greater than the number of geometrically necessary dislocations by
the so-called Nye factor r (∼2, Swadener et al., 2002), thus, the total dislocation density is:

ρT = r ρ G + ρ S (3)

where ρ S is the statistically stored dislocation density. Previous works (Nix and Gao, 1998; Swadener
et al., 2002) assume ρT related to the shear strength τ p by the Taylor (1938) hardening model, i.e.,
τ p = αµb ρ T , where µ is the shear modulus and the geometrical constant α is usually in the range
0.3-0.6 for FCC metals (Wiedersich, 1964). However we note here that this law cannot be considered
reasonable at the nanoscale, where τ p must approach the nanoscale material strength τ (pnano ) , that, at the
truly atomic scale, must be coincident with the theoretical material strength. Thus, the natural
generalization of the Taylor’s law is straightforward:

2
αµb ⎛ αµ ⎞
τp = , β = ⎜ (nano ) ⎟ (4)
ρ T−1 + β b 2 ⎜τ ⎟
⎝ p ⎠

The correct interpretation of eq. (4) is the definition of the real (thus finite) total dislocation density ρT/
as ρ T/ −1 = ρ T−1 + β b 2 , that, as a consequence of the quantized nature of matter, must have an upper bound
of the order of b −2 , as for a pure single dislocation. This is also reflected in the expression of β ,
noting that αµ is of the same order of magnitude of the theoretical material strength. According to this
interpretation the Taylor’s law survives, but considering the substitution ρ T → ρ T/ for accounting the
discrete nature of the energy dissipation. Note the analogy with quantized fracture mechanics (Pugno
and Ruoff, 2004; Pugno 2006a), that quantizing the crack advancement, as must be at the nanoscale,
predicts a finite theoretical material strength, in contrast to the result of the continuum-based linear
elastic fracture mechanics (Griffith, 1920).
Assuming that the flow stress is related to the shear strength by the von Mises’ rule, i.e.,
σ p = 3τ p , and the hardness to flow stress by the Tabor (1951) factor of 3 (Nix and Gao, 1998;
Swadener et al., 2002), the model gives the following hardness prediction:
 H (S V ) ⎛ δ 2 − 1 H (S V ) ⎛ δ 2 − 1
−1 2 12
⎞ ⎞ H
= ⎜⎜ + 1⎟⎟ , = ⎜⎜ 2 + 1⎟⎟ , δ = nano (5)
H nano ⎝ l S V +1 ⎠ H macro ⎝ δ V (lS ) + 1 ⎠ H macro

with H nano ≡ H (lS V → ∞ ) = 3 3 β αµ (at the atomic scale expected to be coincident with the
3 3αµb
theoretical material hardness), H macro ≡ H (lS V → 0 ) =
r
and l = , i.e., a characteristic
ρ S−1 + β b 2 ρSb
length governing the transition from the nano- to the macro-scale. The two equivalent expressions in
eq. (5) correspond to a bottom-up or a top-down view, even if the bottom-up law is perhaps more
physical, starting from the intrinsic material property H nano . Eq. (5) is the mentioned universal law for
nanoindentation, that provides the hardness as a function only of the ratio between the surface
throughout the energy flux propagates and the volume where the energy is dissipated; or, let simply
say, as a function of the surface over volume ratio of the domain in which the energy dissipation
occurs. Such a law can be applied in a very simple way to treat any interesting indenter geometry.
However, to make a comparison, let us focus on the axially symmetric profiles, i.e., h = h(r ) ,
investigated by Swadener et al. (2002).
Conical indenter. Considering a conical indenter with corner angle ϕ , h(r ) = tan ((π − ϕ ) 2 )r ;
3 tan 2 ((π − ϕ ) 2)
geometrically we found S V = , that introduced into eq. (5) gives
2h

H cone (h,ϕ ) = H macro 1 +

(
1 − δ − 2 )h * (ϕ )
, with h * (ϕ ) = 3 2l tan 2 ((π − ϕ ) 2 ) . For h h * → ∞ or ϕ → π
h + δ h (ϕ )
−2 *

H cone → H macro , whereas for h h * → 0 or ϕ → 0 , H cone → H nano ; only for the case of δ → ∞ ,
H cone = H macro 1 + h* h as derived by Nix and Gao (1998). Note that such a scaling law was
previously proposed by Carpinteri (1982) for material strength (with h structural size). But the
comparison here is not very significant, since the indentation size-effect is not a true size-effect.
Parabolic (spherical) indenter. Consider the case of a parabolic indenter with radius at tip R, i.e.,
h = r (2 R ) , that for not too large indentation depth corresponds also to the case of a spherical
2

indenter; geometrically we found S V = 1 R , that introduced into eq. (5) gives

H parabola (R ) = H macro 1 +
(1 − δ )R
−2 *
, with R * = l ; thus, the hardness should not be a function of the
R +δ R −2 *

indentation depth h. For R R * → ∞ , H parabola = H macro , whereas for R R * → 0 , H parabola = H nano ; only
for the case of δ → ∞ , H parabola = H macro 1+ R * R , as derived by Swadener et al. (2002). This law
describes a true size-effect and agrees with the Carpinteri’s law (1982).
Flat indenter. Consider the case of a flat indenter of radius a, i.e., h = δ (r − a ) ; geometrically we
2πah + πa 2
found S V = , that introduced into eq. (5) gives the expression of H flat (a, h ) . For a l → ∞ ,
2 3πa 3
H flat → H macro , whereas for a l → 0 , H flat → H nano ; interestingly for h l → ∞ H flat → H nano ,
showing an inverse indentation size-effect, in agreement with the discussion by Swadener et al. (2002)
and with the intuition (the contact area does not change when the penetration load or depth increases),
see Ashby (1970). This suggests a new intriguing methodology to derive the nanoscale hardness of
materials by a macroscopic experiment, using large flat punches (but the finite curvature at the corners
is expected to affect the results). On the other hand, for h l → 0 (or h ∝ a ) and δ → ∞ ,
H flat = H macro 1 + a * a , with a * = 3 2 l . This case was only discussed by Swadener et al. (2002), due
to the complexity in their formalism to treat such a cuspidal geometry. Note that the last size-effect law
is again coincident with that of Carpinteri (1982).
The generality of the formulation suggests us that the law of eq. (5) can have large applications.
One example is in the design of nanosyringes, e.g., composed by a single walled carbon nanotube, for
which the force is predicted to be Fnano ≈ πDsH nano , where D is the nanotube diameter and s is the
syringe

wall thickness; accordingly, the maximum length to avoid elastic instability of the nanosyringe will be
lmax ≈ π EI Fnano , where E is the Young’s modulus and I is the moment of inertia of the nanotube.
syringe

For example, considering the following reasonable parameters D=10nm, s=0.34nm, H nano = 10GPa ,
E = 1TPa , would result in Fnano ≈ 107 nN and lmax ≈ 111nm . An additional example is on the design
syringe

of bullet-proof rackets: the work needed to penetrate into the material for a length l is
l
W = ∫ A(h )H (h )dh = l AH , where A is the cross-section bullet projected area; since W must be equal
0

K
to the bullet kinetic energy K, its minimum thickness l is predicted to be lmin = . For example,
AH
considering a bullet mass of 100g, velocity of 1Km/s, A=1cm2 and H=10GPa, would correspond to
lmin ≈ 5cm .
Note that defining the impact strength as the energy spent over the removed volume, the last
example shows that such a parameter is for plastic materials of the order of the hardness, but note: at
the investigated size-scale (and H nano >> H macro ). A similar result is found for brittle materials at the
macroscale, where the impact strength is found to be of the order of the material macro-strength
(Pugno, 2006b): thus, our argument suggests its validity at all the size-scales by considering the related
material strength. Summarizing for plastic material the impact strength is of the order of the hardness
whereas for brittle material of the material strength, but at the considered size-scale. This finding could
have interesting applications also in impact or explosion and tribological studies, from the nano- to the
macro-scale. Other examples of applications could be done; but let us return to the nanoindentation
problem.
Swadener et al. (2002) compared their model with experiments in annealed iridium, using
spherical indenters of different radii ( R ≈ 14, 69, 122, 318, 1600 µm ). Data were analyzed using the
Oliver and Pharr (1992) method. Swadener et al. (2002) treated the spherical indentation with their law
for parabolic indenter, since those experiments were performed at a small value of penetration
( a R ≈ 0.05 ). A deviation for R < 80µm , i.e., a lower hardness with respect to their prediction, was
experimentally observed. This deviation is in agreement with the prediction of eq. (5). In particular,
Swadener et al. (2002) considered two plausible sets of parameters for describing their experiments
(see their motivations for details): (a) H macro ≈ 0.9GPa , R * ≈ 250µm or (b) H macro ≈ 0.6GPa ,
R * ≈ 750µm . Introducing such values into our model with in addition (a) H nano ≈ 3GPa or (b)
H nano ≈ 5GPa result in a closer agreement between theory and experiments, as shown in Figure 2 (a)
and (b) respectively. Similar results with the same sets of parameters were observed for pyramidal
indenter (Berkovich, treated as a conic; from its geometry R * h * ≈ 5.2 , Swadener et al. 2002) on the
same material, by varying the indentation depth. Annealed oxygen-free copper tested with spherical,
Berckovich and Vickers (pyramidal, R * h * ≈ 5.2 ) indenters resulted in H macro ≈ 0.1GPa and
R * ≈ 200µm , but with the expected deviation at the smaller size-scales, again well-predicted by our
model with H nano ≈ 2GPa . The same but cold-worked material was similarly investigated, giving
H macro ≈ 0.9GPa and R * ≈ 3.6µm , with a deviation again well described by the finite value
H nano ≈ 2GPa .
Furthermore, we note that at the truly nanoscale we expect even higher values for H nano . A
tendency to very high values, observed by a reduction in the classical slope -1/2, has been observed in
the indentation hardness of surface Si(111) films with an indentation depth as small as 1 nm (Bhushan
and Koinkar, 1994). The hardness for an indentation depth of 2.5 nm is 16.6 GPa and drops to to 11.7
GPa at a depth of 7 nm, from which we deduce a slope of -0.34 (and thus H nano > 20GPa ) .
However, our treatment is not restricted to nanoindentation. Infact, for a given nominally constant
property P the generalization of eq. (5) is straightforward:

−1 γ
⎛ (Pnano Pmacro )γ − 1 ⎞
P(S V ) = Pnano ⎜⎜ + 1⎟⎟ (6a)
⎝ l S V +1 ⎠

in which the parameter γ has to be introduced, since a material property could also for example be
P = H 2 (for which γ = 1 , to match eq. (5)) or P = H −1 ( γ = −2 ), or others. For steady state energy
flux S and V are per unit time. The asymptote at Pmacro is classical, and conceptually intrinsic in
considering a nominally constant property, whereas that at Pnano appears as a consequence of the
existence of a nanoscale quantization. The law of eq. (6a) can be applied for predicting the scaling of a
given property, starting from the surface over volume ratio of the domain in which the energy exchange
or flux, not necessary a dissipation, occurs. For material strength P = σ , γ = 2 (as for H), and
considering self-similar structures, i.e., V S ∝ L , as the characteristic structural size (but note that in
l*
general eq. (6a) describes also the shape-effects), we deduce σ (L ) = σ macro 1 + with
L + l/
l * = (1 − δ −2 )l , l / = δ −2 l and δ = σ nano σ macro (smaller is stronger). This is a nano-scaling law, taking
into account the quantization of the energy flux, and for δ → ∞ agrees with the Carpinteri’s law (1982)
(correspondence principle), that has already been demonstrated to agree with microtorsion (Fleck et al.,
1994) and microbend (Stolken and Evans, 1998) experiments, see Gao et al. (1999a,b) and Huang et al.
(2000). An infinity of other examples could be mentioned, e.g., scaling of fracture energy, friction
coefficient, wear resistance, elastic modulus, and others, to cite a few in the material science.
The law of eq. (6a) can be applied also to complex and chaotic systems, where the multiscale
energy flux arises in a fractal domain of positive dimension D (usually comprised between 2 and 3, i.e.,
between an Euclidean surface and volume); in this case S ∝ V D 3 (Carpinteri and Pugno, 2002), no
matter if we are considering fractal fragments or dislocations, thus brittle or plastic materials
(Carpinteri and Pugno, 2005). Accordingly, S V ∝ LD −3 in eq. (6a) with L structural size. For example,
for a hierarchical material (as for bone and dentine) we derive D = 3 ln n ln (n ϕ ) , where n > 1 and
0 < ϕ < 1 are the number and volumetric fraction content of sub-inclusions in a main inclusion (the
demonstration is left up to the reader). One practical example is given by the scaling of the energy
density P = ψ (nominally a material constant) during fragmentation of solids, for which γ = 1 (since
ψ ∝ H 2 ), in agreement at the intermediate size-scales with the mesoscopic scaling ψ ∝ S V ∝ LD −3
(Carpinteri and Pugno, 2002). Interestingly, such a law is also extensively applied for describing the
scaling of the energy per unit mass spent by biological systems on growth (West et al., 2004; Delsanto
et al., 2004; Carpinteri and Pugno, 2005): thus, eq. (6a) straightforwardly extends this biological
scaling law, as well as the large number of allometric biological laws that can be derived from it (West
et al., 2004). An infinity of other applications could be mentioned, as the scaling of the efficiency of
nanoparticle-based rocket propellants, (fractal particle distribution), of absorption or corrosion and of
other surface properties.
The last application is in the field of fast and slow dynamics (Delsanto et al., 2005). For a wave
(on in general signal) propagation the size-scale L is connected to the time-scale t by L ∝ t (smaller is
faster); in this context eq. (6a) is rewritten as ( S V ∝ 1 L ∝ 1 t ):

⎛ (P P )ν − 1 ⎞
−1 ν

P(t ) = Pfast ⎜ fast slow + 1⎟ (6b)

⎜ τ t + 1 ⎟
⎝ ⎠

where τ is a characteristic time.

An example for the material science community is on the variability of the dynamic strength
P = σ as a function of the time to failure t. The impact strength ( σ fast ) is observed for the majority of
the systems to be approximately twice the static strength ( σ slow ); this seems to be related to the
existence of an incubation time ( τ ) for fracture nucleation, of the order of the time needed to generate
a fracture quantum and thus again related to a quantization (Pugno, 2006a). It is evident that eq. (6b)
with σ fast ≈ 2σ slow catches the described phenomenon (quantized dynamic fracture mechanics would
suggest ν = 1 2 , see Pugno, 2006). The Young’s modulus has a similar scaling as a consequence of
visco-plastic activation, but an infinity of other applications are straightforward, as on the conditioning
and frequency shift in the fast and slow dynamics (Delsanto et al., 2005) or earthquakes triggering,
similar to an crack propagation incubation time, but arising at the megascale. For complex and chaotic
fractal systems, the result previously reported ( S V ∝ LD −3 ) is formally traduced in eq. (6b) as
τ ∝ t D −2 , where D is here connected to the fractal dimension of the time distributions.
To formulate an even more general universal spatial-temporal scaling law P (S V , t ) we have to
combine eqs. (6a) and (6b); two different ways can be followed: considering Pnano and Pmacro in eq. (6a)
as a function of time and of Pnano , Pnano , Pmacro , Pmacro according to eq. (6b) or, complementary,
fast slow fast slow

considering Pfast and Pslow in eq. (6b) as a function of the surface over volume ratio and of Pfast , Pslow ,
nano nano

Pfast , Pslow according to eq. (6a); note that Pfast ≡ Pnano and so on, thus, synthetically, the condition of
macro macro nano fast

symmetry Pab ≡ Pba holds. Following the two different approaches we find the same result, as must be
for self-consistency if and only if δ = Pslow Pslow = Pfast Pfast , i.e., synthetically,
nano macro nano macro

Pab Pac = Pdb Pdc ≡ Pb Pc . Accordingly:

 ⎛ (Pfast Pslow )ν − 1 ⎞
−1 γ −1 ν
⎛ (Pnano Pmacro )γ − 1 ⎞
P(S V , t ) = Pnano ⎜⎜ + 1⎟⎟ ⎜ + 1⎟ (6c)
fast ⎝ l S V +1 ⎜ τ + ⎟
⎠ ⎝ t 1 ⎠

Eq. (6c) is the spatial-temporal universal scaling law that we propose; note that accordingly to the self-
consistent condition we require in addition Pab = Pac Pb Pc ; thus, all the limit cases of eqs. (6a) and (6b)
are recovered. The generalization of eq. (6c) to include new parameters in addition to S V and t (e.g.,
a velocity) is evident. One example of application is on the friction coefficient, for which a spatial-
temporal scaling is observed (“fast” and “slow” here would define the “static” and “dynamic” friction
coefficients); note that our approach quantifies also the influence of the indenter shape (and not only of
its size), e.g., on the friction coefficient.
Concluding, the universality of the derived spatial-temporal scaling could have large applicability
in different fields, even for complex, chaotic and fractal systems, as present in engineering, physics,
biology, medicine, economy, to cite a few; and obviously in nanoindentation.

Acknowledgements
The author would like to thank Profs. A. Carpinteri and P. P. Delsanto for discussion and Renee Eaton
for the English grammar supervision.

Mott, B.W., 1956. Micro-indentation Hardness Testing. Butterworths, London.

Ashby, M.F., 1970. Philos. Mag. 21, 399–424.
Nye, J.F., 1953. Acta Metall. 1, 153–162.
Nix, W.D., Gao, H., 1998. J. Mech. Phys. Solids 46, 411–425.
Poole, W.J., Ashby, M.F., Fleck, N.A., 1996. Scripta Mater.34, 559–564.
Swadener, J.G., George, E.P., Pharr, G.M., 2002. J. Mech. Phys. Solids 50, 681–694.
Lim, Y.Y., Chaudhri, M.M., 1999. Philos. Mag. A 79, 2879–3000.
Arsenlis, A., Parks, D.M., 1999. Acta Mater. 47, 1597–1611.
Taylor, G.I., 1938. J. Inst. Met. 13, 307–324.
Wiedersich, H., 1964. J. Met. 16, 425–430.
Pugno, N., Ruoff, R., 2004. Philos. Mag., 84/27, 2829–2845.
Griffith, A.A., 1920. Phil. Trans. Roy. Soc. A, 221, 163–198.
Tabor, D., 1951. The Hardness of Metals. Clarendon Press, Oxford.
Oliver, W.C., Pharr, G.M., 1992. J. Mater. Res. 7, 1564–1583.
Carpinteri, A., 1994. Int. J. Solids Struct. 31, 291–302.
West, G. B., and Brown, J. H. 2004. Phys. Today 57, 36–43.
Fleck, N.A., Muller, G.M., Ashby, M.F., Hutchinson, J.W., 1994. Acta Metall. Mater. 42, 475–487.
Stolken, J.S., Evans, A.G., 1998. Acta Mater. 46, 5109–5115.
Gao, H., Huang, Y., Nix, W.D., 1999a. Naturwissenschaften 86, 507–515.
Gao, H., Huang, Y., Nix, W.D., Hutchinson, J.W., 1999b. J. Mech. Phys. Solids 47, 1239–1263.
Carpinteri, A., Pugno, N., 2002. J. of App. Mech. 69, 854-856.
Carpinteri, A., Pugno, N., 2005 (June). Nature Mat., 4, 421–423.
Delsanto, P.P., et al., 2004. Appl. Phys. Lett. 85, 4225–4230.
Delsanto, P.P., The Universality of Nonclassical Nonlinearity, Springer, 2005, in press.
Pugno, N., 2006. Int. J. of Fracture, in press.
Pugno, N., 2006. J. of Applied Mech., in press.
Bushan, B., Koinkar, V.N., 1994. Appl. Phys. Lett. 64, 1653–1655.
 FIGURES

a
h
V

Figure 1: Geometrical necessary dislocations during indentation: h is the indentation depth, a is the
radius of the projected contact area, S is the contact surface, and V is the dissipation domain
(proportional to a3). Note that the indented surface at the nanoscale appears in discrete steps due to the
formation of dislocation loops, i.e., of quantized plasticity. In our model the scaling law is predicted to
be a function only of S/V.
 5
4.5 Experiments
4 Micro (previous) - model
3.5 Nano (present) - model

H [GPa]
3
2.5
2
1.5
1
0.5
0
1 2 3 4
log(R /µm)

(a)

4.5
4 Experiments
3.5 Micro (previous) - model
3 Nano (present) - model
H [GPa]

2.5
2
1.5
1
0.5
0
1 2 3 4
log(R /µm)
(b)

Figure 2: Comparison between “micro-models” (Nix and Gao, 1998 and Swadener et al., 2002) and
present “nano-model” fitted to experiments on spherical indentation by Swadener et al. (2002). They
considered two different sets of parameters: (a) H macro ≈ 0.9GPa , R * ≈ 250µm (dotted line), that
introduced in the nano-model (solid line) with H nano ≈ 3GPa result in a closer agreement with the
experiments (points); (b) same, but for H macro ≈ 0.6GPa , R * ≈ 750µm (dotted line), or in addition
H nano ≈ 5GPa (solid line). Note the difference in the predictions of the models for the smaller size-
scale.