Supporting Information for the paper entitled: “Evidence of a fractal percolating network during geopolymerisation” by J. Rouyer and A. Poulesquen I - Theoretical background 1. Oscillatory Rheology (OR) Dynamic rheology is an approach which is widely used to measure viscoelastic moduli. Taking a fixed pulsation ð0 , a sinusoidal monochromatic strain is applied to the sample under investigation and the resulting harmonic stress response is recorded. The stress ð and the strain ðū are expressed as follows in the linear regime: ðū(ðĄ) = ðū0 sin(ð0 ðĄ) { ð(ðĄ) = ð0 sin(ð0 ðĄ + ðŋ) (1) where ð0 and ðū0 are the stress and the strain magnitude, respectively; the so-called loss angle ðŋ is the loss angle between the strain and stress signals. The viscoelastic moduli ðš′ and ðš′′, are classically defined by: ð0 cos ðŋ ðū0 { ð0 ðš ′′ (ð0 ) = sin ðŋ ðū0 ðš ′ (ð0 ) = (2) Stress and strain magnitude and the loss angle can either be calculated directly from the temporal signals or obtained by performing calculations in the angular frequency domain. ð0 and ðū0 are the spectral weight of the strain and stress Fourier transforms at the given pulsation ð0 . Likewise, the loss angle can be obtained by calculating the difference between the phase angles in the strain and stress phases. Calculations in the temporal and angular frequency domains yield strictly identical results. When dynamic rheology methods are extended to a multi-decade angular frequency range, it becomes a powerful mean of describing the rheological regime and the mechanisms occurring between the components of the material under investigation. By varying the angular frequency within the scope of the rheometer bandwidth, it is possible to draw up the viscoelastic spectrum of the sample. Elastic versus viscous moduli evolution can be used to describe the rheological behavior of the material using well-known criteria such as the slope values, the parallelism or the crossover between the curves of the moduli1; in this way, the evolution of the viscoelastic state, the state of the gel and the relaxation times can be determined. Successive iterations of several monochromatic angular frequencies within the useful range can be performed during a relatively long period of time, depending on the degree of refinement required. This duration will be at least equal to 182 minutes; for example, in the case of four points per decade at angular frequencies ranging from 0.01 to 100 rad/s and a 10-cycle strain sine per angular frequency. In order to obtain an accurate viscoelastic spectrum, the sample must show stable properties throughout this time. Dedicated controlled environmental tools such as thermal devices, solvent traps and convection ovens provide helpful means in order to maintain a consistent steady state. However, this procedure is not at all suitable when spontaneous chemical reactions or structural adjustment processes are liable to occur. More specifically, the typical time taken by internal changes to occur has to be infinite in comparison with the duration of the experiment. In the case of evolving materials, another means of approaching this problem could consist of changing the mathematical formulation of the harmonic strain load. 2. Time-Frequency-Resolved Rheology (TF2R) This method, which was firstly described by Ghiringelli et al2, consists in using a relatively short strain load, which contains an extended continuous angular frequency range. It differs from the multi-wave technique3,4, where strain function is obtained by summating a few monochromatic sine waves, resulting in a sparse spectrum (i.e. one pair of parameters at each monochromatic frequency). The main advantage of the TF2R method is that it can be used to measure a continuous spectrum in a multidecade frequency range experimentally in a relatively short time. In order to illustrate how this approach is used, the frequency content of some strain functions, which are commonly used in rheology, was studied from the signal processing point of view. A peak holding experiment can be said to consist theoretically of a Dirac function. By definition, the frequency content of this punctate function is homogeneous in the Fourier domain5, as shown in Figs. 1 and 2, which present the temporal and frequency curves, respectively. From the practical point of view, these strains are not compatible with the experimental constraints, which involve sampling limitations (maximum sampling frequency ðð =1000 Hz) and quite-long spinup times due to the mechanical and geometrical inertia. To overcome these limitations, a suitable square function or Gaussian function has to be chosen in order to smooth out the beginning and the end of the rotation (as in the example presented in Fig. 1). The theoretical advantage of applying an extremely short strain load corresponding to an infinite frequency range amounts in practice to having a fairly narrow bandwidth. As illustrated in Fig. 2, the square function and Gaussian function spectra give a homogeneous bandwidth up to an angular frequency of 3 rad/s. Otherwise, higher angular frequencies cannot be reached with these functions. In other words, the experimental device acts like a low-pass filter when dealing with extremely short strains. In order to span all the frequencies covered by the rheometer bandwidth in a relatively short time, the use of the chirp function was found to provide the best solution. Here we focus on a four-decade bandwidth ranging from angular frequencies of 0.01 to 100 rad/s. A chirp is a pseudo-harmonic function obtained by applying a continuous frequency sweep with time. In order to maximize the signal-to-noise ratio at the lower frequencies, an exponential frequency sweep was used, which was expressed as follows: ð ðū(ðĄ) = ðū0 sin ( ðū1 (ð ðūðĄ − 1)), with ðū = ð ∈ [ð1 , ð2 ], ð ln 2 (3) ð1 ð1 The angular frequency sweep, which starts from ð1 to ð2 . ð1 , corresponding to the duration of the sweep, is classically equal to the period of the lowest angular frequency, and hence ð1 = 2ð ⁄ð1 , where ðū is the sweeping rate of the strain and ðū0 is the magnitude of the strain. An example of a ð1 = 628 seconds chirp is shown in Fig. 3. The frequency distribution is continuous over the bandwidth of interest, as shown in Fig. 2. The spectral weight of the chirp is always greater than the other loads involved. It is worth noting that the magnitude of the strain is the same in all the temporal functions. The energy contributed by each frequency is greater than with the other short loads tested such as Dirac, square and Gaussian functions. However, a typical noisy oscillation appears in the chirp spectrum at the upper limit of the high frequencies. In order to obtain an unperturbed bandwidth, the upper boundary is moved to ððĒ = 400 rad/s. In the same way, the lowest angular frequency is decreased to ðð = 10-3 rad/s. The spectrum of the extended bandwidth is plotted in Fig. 2; at the lowest frequencies involved, the spectral weights have been maximized and the bandwidth of interest is devoid of undesirable artifacts. It can be noted that expanding the bandwidth (keeping the same duration) firstly results in a higher sweeping rate and secondly, in a slight overall decrease in the spectral weight. In the general case of linear viscoelastic materials, the stress response is defined by the convolution product (⊗) of the complex shear modulus and the strain velocity: ð(ðĄ) = ðš(ðĄ) ⊗ ðūĖ (ðĄ) (4) ðš(ðĄ) is the so-called relaxation function of the sample. The use of the convolution product makes it possible to shift from the temporal domain to the Fourier domain (~), and the previous relation becomes: ðĖ(ð) = ðððšĖ (ð)ðūĖ(ð) (5) The method used to model the viscoelastic moduli, which has been presented in detail in [23], gives the following two expressions: ðĖ(ð) ] ðūĖ(ð) ðĖ(ð) ðš ′′ (ð) = ℑ [ ] ðūĖ(ð) { ðš ′ (ð) = ℜ [ (6) A snapshot of the viscoelastic behavior of the material under investigation can therefore be obtained by calculating the stress-to-strain ratio in the previously defined bounded Fourier domain. 3. Mutation number in the case of TF2R As mentioned in previous section, the possibility to obtain meaningful rheological measurements depends on the properties consistency of the sample during the time required by the measuring apparatus. This experimental prerequisite makes necessary to adapt the duration ð of the chirp to the changes in the properties of the material. The mutation number of 6,7 is therefore introduced in order to predict the changes liable to occur during the pseudo-harmonic loading. In the present case, both moduli will result in the following two dimensionless numbers (ðððĒ ): ð ′ ððĒ (ðĄ) = ð ððš ′ ð ððš′′ and ð ′′ ððĒ (ðĄ) = ′ ðš ððĄ ðš′′ ððĄ (7) The smaller the mutation number is, the more negligible the mutations will be in the analysis of the experimental results. Negligible non-linear effects were observed with small mutation numbers2, and the moduli were accurately determined within the duration ð as long as ðððĒ was smaller than 15% 6; if this finding is confirmed, the sample can be assumed to be quasi-stable during the experiment. One rad/s monochromatic paste setting kinetics (Fig. 1 in the main text) were therefore used to calculate the mutation number during a period of ð1 = 628 seconds. It can be seen from ′ ′′ Fig. 5 that the patterns of evolution of ðððĒ and ðððĒ were very similar: only the mutation number of the elastic modulus was therefore taken into account. The highest value recommended – 0.15 – was greatly exceeded during the process, and the chirp initially included in equation (3) therefore turned out to be unsuitable. In order to make the length of the chirp comply with the recommended value, the duration ð1 was decreased whenever the mutation number was greater than 0.15. Two other chirp durations were therefore chosen: that of the shortest chirp observed, and an intermediate duration, which were defined as follows: ð2 = 368 s and ð3 = 108 s. An experimental procedure integrating the three appropriate durations was defined at specific times during the setting process, as shown in Fig. 5, where they form a stepwise pattern. With a view to investigate the effects of using shorter chirp durations in the TF2R method, preliminary tests were conducted on PDMS samples (Annex A in the main text). These tests led us to reduce the frequency range of the viscoelastic spectrum, depending on the chirp duration. The spectral range adopted was therefore [0.01, 100] rad/s in the case of ð1 , [0.05, 100] rad/s in that of ð2 and [0.1, 100] rad/s in that of ð3 . In addition, the possibility of obtaining reliable moduli at frequencies of less than 0.1 rad/s depends not only on having sufficiently long chirp durations, but even more on the order of magnitude of the torque measurements2. The low sensitivity of the torque transducer can make difficult to obtain the theoretical full-bandwidth characterization expected. II - Testing the TF2R method on a PDMS sample PDMS is a silicon-based organic polymer, which was used as a viscoelastic test material since it is non-evolving and shows excellent long term stability. The viscoelastic spectrum was measured at angular frequencies ranging from 0.01 to 100 rad/s using oscillatory rheological methods at 25°C and a strain magnitude equal to 101%. Four points per decade were determined. Measurements were performed using plate-plate geometry with a diameter of 2.25 cm. TF2R was then carried out with the exponential chirp strain loading at angular frequencies ranging from ðð = 0.001 rad/s to ððĒ = 400 rad/s and ðū0 = 10-1 %. The superimposed viscoelastic spectra obtained using the TF2R method and standard dynamic rheological methods are presented in Fig. 4 in the case of the three chirp durations (ð1 = 628 s, ð2 = 368 s and ð3 = 108 s). The relative error ðŋðš between the standard method and the TF2R method is entered in Table 1. The discrete points in the OR were accurately matched at angular frequencies greater than 0.3 rad/s with a relative error of less than 4 % with all the chirp durations tested; the crossover moduli could therefore be accurately determined. As shown in Table 1 and Fig.6, the longer the chirp duration is, the more accurately this technique can be expected to give fine moduli values at low frequencies. The viscous modulus was accurately assessed in general, but a limitation emerged as regards the description of the elastic modulus at the lower frequencies tested, when the measurements were disturbed by strong oscillations (Fig.6 for ð3 ). This undesirable effect can be attributed to the small number of sine periods present in the lower pulsation range: the lack of redundancy reduces the accuracy of the TF2R method at the lower frequencies. It was therefore decided to reduce the bandwidth, depending on the chirp durations used; [10-2; 102] rad/s in the case of ð1 , [5.10-2; 102] rad/s in that of ð2 and [10-1; 102] rad/s in that of ð3 . Particular attention will therefore have to be paid to how the spectrum is interpreted in the lower angular frequency range. These preliminary experiments provide a useful framework for applying the TF2R method to geopolymerization processes. Table 1 Relative error ðđðŪ between the standard method and the TF2R method with the three chirp durations (ðŧð = 628 s, ðŧð = 368 s and ðŧð = 108 s) to determine the viscoelastic moduli. Pulsation 0.01 0.03 0.13 0.32 0.80 2.00 5.02 12.62 31.70 79.62 ð ðŪ′ for ðŧð [%] (-) (-) -13.56 2.08 1.44 0.63 3.08 3.00 2.87 2.40 ð ðŪ′′ for ðŧð [%] (-) -30.42 0.22 1.58 3.59 2.17 2.41 2.58 3.46 2.16 ð ðŪ′ for ðŧð [%] (-) 30.77 2.92 0.37 3.20 1.15 2.60 2.42 2.44 1.57 ð ðŪ′′ for ðŧð [%] -8.75 3.45 1.17 2.03 2.41 2.43 2.95 3.49 3.95 5.04 ð ðŪ′ for ðŧð [%] -86.89 -1.33 0.16 2.44 3.23 1.36 1.93 2.18 2.47 3.09 ð ðŪ′′ for ðŧð [%] -0.93 -0.69 1.18 2.05 2.57 1.89 2.90 3.16 3.82 1.42 [rad/s] Relative errors were calculated using: ðŋðš = 1 − ðšððđ2ð /ðšðð – The symbol (-) denotes aberrant values in the TF2R results. Modulus values tended to diverge at low frequencies, as shown in Fig.A. Supporting information references 1. Macosko CW. Linear Viscoelasticity. In: Rheology: Principles, Measurements, and Applications. Wiley-VCH; 1994:109-126. 2. Ghiringhelli E, Roux D, Bleses D, Galliard H, Caton F. Optimal fourier rheometry. Rheol. Acta 2012;51(5):413-420. 3. Ma J, Lin Y, Chen X, Zhao B, Zhang J. Flow behavior, thixotropy and dynamical viscoelasticity of sodium alginate aqueous solutions. Food Hydrocoll. 2014;38(0):119128. 4. Nachbaur L, Mutin JC, Nonat A, Choplin L. Dynamic mode rheology of cement and tricalcium silicate pastes from mixing to setting. Cem. Concr. Res. 2001;31(2):183192. 5. Smith DC. An introduction to distribution theory for signals analysis.: Part II. The convolution. Digit. Signal Process. 2006;16(4):419-444. 6. Winter HH, Morganelli P, Chambon F. Stoichiometry effects on rheology of model polyurethanes at the gel point. Macromolecules 1988;21(2):532-535. 7. Mours M, Winter HH. Time-resolved rheometry. Rheol. Acta 1994;33(5):385-397. doi:10.1007/bf00366581. 6. Winter HH, Morganelli P, Chambon F. Stoichiometry effects on rheology of model polyurethanes at the gel point. Macromolecules. 1988;21(2):532-535. 7. Mours M, Winter HH. Time-resolved rheometry. Rheol Acta. 1994;33(5):385-397. Supporting information figures Fig. 1 Normalized graphs of the Dirac delta function (black dotted circle), the 1-second wide square function and the 0.425-standard deviation Gaussian function. Fig. 2 Spectral curve of the Dirac delta, the square function, the Gaussian function and the exponential chirp functions used to illustrate the interest in the TF2R method. These curves correspond to the module of the Fourier transform of the normalized function. This qualitative comparison shows that the chirp gives a large spectral weight within a wide pulsation range. Fig. 3 Representation of the exponential chirp defined with the following parameters: ð1 =0.01 rad/s and ð2 =100 rad/s and ð1 = 628 seconds. The corresponding spectral curve is shown in Fig. 2 (thin continuous line). Fig. 4 Evolution of the mutation numbers calculated with the 1-rad/s viscoelastic moduli measured during the setting process (Fig.4) and calculated using equation (7). The values plotted correspond to an experimental time of T = 628 second, which is the duration of the exponential chirp defined by equation (3) and shown in Fig.3. Mutation numbers have to be less than 0.15 in order to obtain reliable measurements within this time; this figure illustrates the situation where the experimental duration of the chirp had to be decreased. Fig. 5 Analysis of the mutation number as a function of the moduli led to adapting the chirp duration to obtain a mutation number smaller than 0.15. First the shorter duration was found to be equal to 108 seconds (a median-long duration). The chirp duration was then adapted to the threshold meaningful mutation number, which was set at 15%. This fig. gives the chirp durations chosen during the geopolymerization process, depending on the increase in the moduli. Fig. 6 Comparison between the standard OR and the TF2R methods. A 108 (a), 368 (b) and 628 seconds (c) chirp duration was employed to measure the continuous viscoelastic spectra by TF2R (plain and dashed lines). The OR determines moduli iteratively at discrete pulsation (squares and circles) and takes about 182 minutes to obtain 4 points per decade within a four decade pulsation range.