-_ 7 cf) Impulsive Stimulated Thermal Scattering of Glass-Forming Liquids by Dora Marie Paolucci B.S. (Chemistry and Mathematics) University of Richmond (1993) Submitted to the Department of Chemistry in partial fulfillment of the requirements for the degree of Doctor of Philosophy at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY June 1998 @ Massachusetts Institute of Technology 1998. All rights reserved. Auth or .. .... , Certiified by ............. 0. my Department of Chemistry 21, 1998 I rMay .................................... Keith A. Nelson Professor of Chemistry Thesis Supervisor Accepted by ..... oiF. r,-i TTS oH!A INSTiJ'I OF rECHt VI ': JiON 1.51998 FOL6 // Dietmar Seyferth Chairman, Departmental Committee on Graduate Students This doctoral thesis has been examined by a committee of the Department of Chemistry as follows: Professor Irwin Oppenheim 1 Chairman Professor Keith A. Nelson Thesis Supervisor Professor Robert Silbey 7?. Impulsive Stimulated Thermal Scattering of Supercooled Liquids by Dora Marie Paolucci Submitted to the Department of Chemistry on May 21, 1998, in partial fulfillment of the requirements for the degree of Doctor of Philosophy Abstract Impulsive Stimulated Thermal Scattering is a time domain laser spectroscopy that is applied to studying the relaxation dynamics of liquids in the supercooled temperature regime. Characterizing the relaxation dynamics of supercooled liquids is essential to developing an understanding of the physics of the liquid-glass transition. Mode Coupling Theory is a popular theory that has had some success in explaining experimental observations of the dynamics of certain supercooled liquids. Glass forming liquids are categorized according to fragility. To date, many of the experimental and theoretical studies of glass forming liquids have focused on liquids with fragile characteristics. To extend the field of research, ISTS studies are conducted on glycerol, a glass-forming liquid of intermediate strength. Other experimental studies of glycerol have given conflicting results, and the applicability of Mode Coupling Theory is questioned. ISTS is added to the other experimental techniques to measure the DebyeWaller factor and relaxation times in the regime from a few nanoseconds to hundreds of microseconds. There is no MCT anomaly in the temperature range measured. Also, the shape of the relaxation spectrum is constant, without the enhanced stretching commonly observed at the crossover temperature Tc. ISTS is also applied to characterization of the dynamics of a glass-forming polymer, Polypropylene Glycol. Like many polymers, Polypropylene Glycol shows two features in the slow relaxation portion of the dielectric spectrum. In ISTS, only the feature which is strongly coupled to the density is observed, at least quantitatively. A comparison of the results for two molecular weights shows that the relaxation observed in ISTS is independent of molecular weight, and therefore related to the main a peak in the dielectric spectrum, which is also independent of molecular weight. Some temperature dependence is observed in the Debye-Waller factor and the stretching of the relaxation spectrum. There is no Debye-Waller factor anomaly as predicted by MCT in the temperature range measured. The conclusion reached from ISTS experiments on non-fragile glass forming liquids is that, despite the success MCT has enjoyed in predicting the behavior of relatively simple systems, there is not sufficient detail incorporated in the theory to address the complexity of networked materials (glycerol) and polymers. Thesis Supervisor: Keith A. Nelson Title: Professor of Chemistry Acknowledgments I would like to thank my advisor, Keith Nelson, for his constant enthusiasm for doing science. He is always a source of valuable advice, and always believes a problem can be solved. I am very thankful for the respect he has shown me. I would also like to thank my undergraduate advisor, Sam Abrash, for teaching me physical chemistry and starting me on the research path. The enthusiasm of Mr. Elmer and Mr. Moran, my high school chemistry and biology teachers, is largely responsible for my interest in science. Over the years, I have been privileged to work with a wide variety of people. I owe Laura Muller many thanks for introducing me to the world of lasers and liquids. She is a remarkably patient mentor, and always understands that sometimes you just need to "walk like an Egyptian". This research on supercooled liquids has greatly benefited from the work of earlier students in the Nelson group. In particular, I would like to acknowledge Yongwu Yang for his work on the hydrodynamic treatment of ISTS, and the experiments on fragile liquids. Many of the figures in Chapter 4 are the results of his work. Rebecca Slayton has provided essential help and encouragement in this last year, and will be continuing the experimental efforts in this field. Tanya, Alex, Ariya, and Randy have also been helpful picosecond lab collaborators. Ciaran Brennan has given invaluable advice about an astonishing number of lab-related things, and has been a good friend as well. I would also like to thank Lisa, John, John, Marc, Dutch, Greg, Richard, and Tim for all their help. A great many people have passed through the Nelson Group and the other groups in the basement of building 2, making it an exciting place to work, and I thank them all. When I started graduate studies at MIT, I was very fortunate to have a wonderful group of classmates. Many times, we donned our helmets and tackled those nasty integrals together, with a spirit of teamwork. Dave Rovnyak in particular has been a good friend and fellow Richmond alum. I wish everyone the best. I am very appreciative of the good and faithful friends who have supported and encouraged me, particularly in this last year. Anand Mehta has always been there for me, through all the ups and the downs. I don't know how I could have done this without him. Amy, Jen, (and Tessa) have been very understanding (and warm and fuzzy) housemates, particularly in these last crazy thesis-writing weeks. Along with the other residents, they have helped make the Berskshire St. house a warm and safe place. I would also like to thank everyone involved in eaps-vball, the Women's Volleyball Club, and the various Chemistry hockey teams for all the good times we have enjoyed. Most of all, I would like to thank my parents, who have given me so much. From the day I ran down "the path" to kindergarten, my parents have always encouraged me to tackle any challenge that came my way. I would like to extend very special thanks to them and the rest of my family for all of their love over the years. Table of Contents List of Abbreviations 9 List of Figures 10 List of Tables 14 Chapterl Introduction 15 Chapter 2 Mode Coupling Theory 20 2.1 Mode Coupling Theory Background 20 2.2 Experimental Methods of Testing Mode Coupling Theory 33 Chapter 3 Impulsive Stimulated Thermal Scattering 36 3.1 Overview and Theory 36 3.2 Experimental 38 3.2.1 Nd:Yag Laser 38 3.2.2 Generation of the Excitation Grating 41 3.2.3 Probing the Grating 46 3.2.4 Detecting the Signal 48 3.2.5 Temperature Control 49 3.2.6 Automation 50 3.3 Heterodyne Signal 51 3.4 Data Analysis Methodology 57 3.4.1 Accounting for Detector Response 57 3.4.2 Time Domain Fits to the Data 66 3.4.3 Alternate Analysis of the Relaxation Mode 66 Chapter 4 Review of Impulsive Stimulated Scattering on Fragile Liquids 68 Chapter 5 Impulsive Stimulated Thermal Scattering on Glycerol 76 5.1 Experimental background 76 5.2 Experimental 81 5.3 Results 82 5.3.1 Acoustics 85 5.3.2 Thermal Diffusivity 91 5.3.3 Relaxation and Debye-Waller factor measurements 94 5.4 Discussion 102 5.5 Conclusions 108 Chapter 6 Impulsive Stimulated Thermal Scattering on Polypropylene Glycol 111 6.1 Overview 111 6.2 Experimental 115 6.3 Results 115 6.3.1 Acoustics 118 6.3.2 Relaxation measurements 125 6.4 Discussion 139 6.5 Conclusions 146 Chapter 7 Summary and Future Directions 148 Appendix A Selected Matlab codes 153 Appendix B Selected References 157 List of Abbreviations ISS ISTS ISBS MCT BS PCS KWW VTF PPG PPG425 PPG4000 salol Impulsive Stimulated Scattering Impulsive Stimulated Thermal Scattering Impulsive Stimulated Brillouin Scattering Mode Coupling Theory Brillouin Scattering Photon Correlation Spectroscopy Kohlrausch-Williams-Watts Vogel-Tamman-Fulcher Polypropylene Glycol Polypropylene Glycol, Molecular Weight 425 Polypropylene Glycol, Molecular Weight 4000 phenyl-salicylate CKN Ca 4 K 6(NO 3) 14 YAG cw RTD T, Tc To, TTF Yttrium Aluminum Garnet Continuous Wave Resistance Thermal Detector Glass Transition Temperature Mode Coupling Theory Crossover Temperature Vogel Temperature List of Figures Figure 1-1: Angell plot illustrating the concept of fragility 16 Figure 2-1: Illustration of the relaxation function and relaxation spectrum as predicted by Mode Coupling Theory. 23 Figure 3-1: Illustration of the optical setup for generation and probing of the grating. 43 Figure 3-2: Phase mask design showing patterns obtained for ISTS, ISRS, and other applications. 44 Figure 3-3: Illustration of a simple imaging system for generation and probing of a grating using a phase mask. 47 Figure 3-4: Comparison of simulated data for ISTS signal with different amplitudes of homodyne and heterodyne contribution. 54 Figure 3-5: Simulated data with a heterodyne component (3.3.2) and fit to homodyne ISTS signal (3.1.3). Data are presented on a linear scale. 55 Figure 3-6: Simulated data with a heterodyne component (3.3.2) and fit to homodyne ISTS signal (3.1.3). Data are presented on a log scale. 56 Figure 3-7: Example of raw ISTS data taken with two different detectors. 58 Figure 3-8: Measured Impulse Response of the Antel detector. 62 Figure 3-9: ISTS data taken with the Antel detector. 63 Figure 3-10: Response of the Hamamatsu Detector to a quasi-cw probe. 64 Figure 3-11: ISTS data taken with the Hamamatsu detector. 65 Figure 4-1: Temperature Dependence of the Relaxation Time in Salol. Includes a comparison to Dynamic Light Scattering Data. 69 Figure 4-2: Temperature Dependence of the Debye-Waller factor in Salol, in the q--0 70 limit. Figure 4-3: Temperature Dependence 3 in Salol. Data obtained from Dynamic Light 71 Scattering is included for comparison. Figure 4-4: Temperature Dependence of the Debye-Waller factor in CKN, in the q->O 72 limit. Figure 4-5: Temperature Dependence of P in CKN. Data obtained from other experimental techniques is included for comparison. 73 Figure 4-6: Temperature Dependence of the Debye-Waller factor in butylbenzene 74 Figure 5-1: ISTS data and fits from glycerol at q = 0.305 tm-'. 83 Figure 5-2: ISTS data and fits from glycerol at q = 1.19 tm-'. 84 Figure 5-3: Speed of sound in glycerol at several wavevectors. 86 Figure 5-4: Acoustic damping in glycerol at several wavevectors. 87 Figure 5-5: Acoustic frequency vs. wavevector at several temperatures. 89 Figure 5-6: Acoustic damping vs. q2 at several temperatures. 90 Figure 5-7: Thermal diffusivity vs. temperature in glycerol. 92 Figure 5-8: Relaxation times (log scale) vs. temperature in glycerol at all wavevectors. 94 Figure 5-9: Relaxation times at all wavevectors, with a best fit to Arrhenius temperature dependence. 95 Figure 5-10: Vogel-Tamman-Fulcher fit to relaxation times (In scale) in glycerol. 96 Figure 5-11: Temperature Dependence of P in glycerol at all wavevectors. 97 Figure 5-12: The relaxation function in glycerol at several temperatures and wavevectors. 99 Figure 5-13: Debye-Waller factor in glycerol for all wavevectors. 100 Figure 5-14: Debye-Waller factor vs. temperature at the two smallest wavevectors, to highlight the data at the lowest temperatures. 101 Figure 5-15: Relaxation times in glycerol fit to a power law with no restrictions. 106 Figure 5-16: MCT power law fit with T, restricted to 225 K. 107 Figure 6-1: ISTS data from PPG4000. The data shows the growth of the structural relaxation feature as the sample cools. 116 Figure 6-2: ISTS data from PPG425, showing the structural relaxation appearing at intermediate temperatures. 117 Figure 6-3: Speed of sound in PPG425 at several wavevectors. 119 Figure 6-4: Speed of sound in PPG4000 at several wavevectors. 120 Figure 6-5: Comparison of the speed of sound in PPG425 and PPG4000. 121 Figure 6-6: Acoustic damping in PPG425. 122 Figure 6-7: Acoustic damping in PPG425 at one wavevector, highlighting the low temperature shoulder. 123 Figure 6-8: Acoustic damping in PPG4000 at one wavevector. 124 Figure 6-9: Log plot of relaxation times in PPG425 obtained by ISTS and other experimental methods. 126 Figure 6-10: Comparison of Relaxation Times for PPG425 and PPG4000. 127 Figure 6-11: Vogel-Tamman-Fulcher fit to relaxation times in PPG425. 128 Figure 6-12: VTF fit to relaxation times in PPG4000. 129 Figure 6-13: Average relaxation time in PPG425 fit to a MCT power law for the temperature dependence. 130 Figure 6-14: Average relaxation times, with data at temperatures above 230 K fit to a MCT power law temperature dependence. 131 Figure 6-15: Stretching Parameter 3 from PPG425 as a function of temperature at several wavevectors. 132 Figure 6-16: Stretching Parameter 1 from PPG425 as a function of temperature at several wavevectors, including measurements from VV and VH Photon Correlation Spectroscopy. 133 Figure 6-17: Stretching parameter P in PPG4000 at several wavevectors. 134 Figure 6-18: The Debye-Waller Factor in PPG425 as a function of temperature. 136 Figure 6-19: The Debye-Waller Factor in PPG425 at different wavevectors. 137 Figure 6-20: Debye-Waller factor measured for PPG4000. 138 List of Tables Table 3-1: Simulated and Fit parameters for ISTS signal with a mixture of homodyne and heterodyne signal. 52 Table 5-1: Experimentally determined exponents for the susceptibility spectrum of glycerol. 79 Chapter 1 Introduction The physics of supercooled liquids and the glass transition has been the subject of a stimulating dialog between theory and experiment in recent years. 1-3 A supercooled liquid is a substance that has been cooled below its freezing point without crystallization. As the material is cooled further, the viscosity dramatically increases, accompanied by a corresponding increase in the characteristic time for relaxation processes. At a certain temperature, the glass transition occurs, where changes in the thermodynamic properties are observed. Despite the change in thermodynamic properties, however, the glass transition is not a traditional first or second order phase transition, but an experimentally defined kinetic transition. So far, no conventional order parameter has been defined to describe behavior near the glass transition. The characteristic time scale for structural relaxation at the glass transition is on the order of many seconds, corresponding to a viscosity of 10"'poise. Supercooled liquids and glasses can be formed from a wide variety of chemical substances. Commonly studied glass forming liquids included molecular liquids such as salol and glycerol, ionic melts such as Ca 4K6(NO3) 4 (CKN), and polymers. Other glasses have been less studied from the theoretical perspective, but have important practical applications, including the common "window" glass, and metallic glasses. Glass forming liquids are often categorized by the temperature dependence of the viscosity through the supercooled regime. The two extremes are termed "strong" and "fragile". A strong liquid displays an Arrhenius temperature dependence of the viscosity. The viscosity of the fragile liquid shows extreme deviation from Arrhenius behavior. Although there are liquids showing continuous behavior between the two extremes, the liquids are generally grouped into the three categories of strong, intermediate, and fragile. Figure 1-1 shows an example of the temperature dependence of the viscosity for a strong, an intermediate, and a fragile liquid. 122 10 K2 0 8E - 6 2 glycerol ui" -2 0.5 0.6 I 0.7 -6 "-8 -10 0.8 0.9 1.0 TV T Figure 1-1: Angell plot showing glass-forming liquids of varying fragility. SiO 2 is a strong liquid, glycerol is an intermediate liquid, and ortho-terphenyl (OTP) is a fragile liquids. The y axes show the viscosity or relaxation time, while the x axis is in terms of the reduced temperature T/T, so all three liquids show the same viscosity at the glass transition. This figure is reproduced from Ref. 4 Fragile liquids have been the most studied to date, particularly salol, CKN, and ortho-terphenyl. Although in general, fragile liquids are more challenging experimentally and prone to crystallization, it is easier to simulate their behavior. Strong liquids are seldom studied experimentally, as the glass transition temperature is often extremely high, and the liquids must be further heated if the goal is to study fast dynamics. Intermediate liquids are gaining more and more attention as the natural extension of the work that has been completed for the fragile materials. Polymers are also gaining more attention from a fundamental perspective, as they form glasses easily, and have viscosity behaviors ranging from fragile to intermediate. Much empirical work had previously been done due to the practical uses of polymers. Despite the dramatic differences in chemical composition, there are several properties or behaviors of supercooled liquids and glasses that are universally observed. The theoretical and experimental efforts to explain and characterizes these dynamics amount to a search for the physical mechanism underlying the dramatic increase in the relaxation time through the supercooled regime. Mode Coupling Theory, a kinetic theory of the glass transition, postulates that non-linear feedback between high wavevector modes creates a bottleneck situation which describes the physics underlying the glass transition. Impulsive Stimulated Thermal Scattering (ISTS) is one of many experimental tools that is employed in the study of the dynamics of supercooled liquids. The current dynamic range of ISTS measurements is about 5 decades, from nanoseconds to milliseconds. The location of this range is intermediate between several light scattering techniques, and includes some of the range covered by dielectric techniques. Most importantly, in several well studied glass forming liquids, ISTS covers the frequency range where central experimental predictions are made. ISTS can test these predictions without extrapolation of frequency or temperature. The main difficulty in studying the glass transition experimentally is characterization of the dynamics of a glass forming liquid over the entire temperature range, spanning the extremely wide range of time scales. As a result, to cover the entire spectrum, the data from different experiments are often combined. Whenever there is a change in the experiment, the nature of the measurement must be reevaluated. In many cases, it is far from clear that two experiments probe exactly the same dynamics in the material. Each experiment has some observable or property that it measures, and will give information on modes that strongly affect the value of that specific property. If a mode is not strongly coupled to the specific observable of the experiment, it will be attenuated, or not observed at all. Often, achieving a complete understanding of each experiment is itself a challenge. One of the recurring difficulties in testing Mode Coupling Theory is that the predictions of the theory focus on the time and temperature dependence of the density of the material. ISTS and neutron scattering are the only techniques that are well understood to directly measure density variations in time or frequency. These techniques do not cover the entire relevant frequency range, so other experiments must be incorporated to completely span the dynamics. As a result, there is always the open question of how well the measurement truly relates to the prediction for the density dynamics. Comparison of ISTS results with results from other experiments may help elucidate the similarities and differences between several techniques. The research included in this thesis has several goals. The first is to measure the density dynamics of a glass forming liquid with intermediate strength, glycerol, and to determine how well Mode Coupling Theory explains the results. This is an extension of previous ISTS studies on fragile liquids. The second goal is to begin work on applying ISTS as a technique for studying polymer dynamics. As appropriate, Mode Coupling Theory predictions are also evaluated for the polymer results. In the course of studying these materials, the opportunity arises to compare and contrast the measurements made with ISTS and other experimental techniques. Chapters 2, 3, and 4 provide background information. Chapter 2 outlines Mode Coupling Theory, experimentally testable predictions, and describes several experimental techniques commonly used to test Mode Coupling Theory. Chapter 3 gives details on the Impulsive Stimulated Thermal Scattering experiment and data analysis. Chapter 4 briefly reviews some data obtained from ISTS from fragile glass forming liquids. The results of ISTS studies on non-fragile glass forming liquids, glycerol and Polypropylene Glycol, are presented in Chapters 5 and 6. Chapter 5 presents the experimental results for glycerol and a Mode Coupling Theory analysis of the results. Two molecular weights of Polypropylene Glycol are studied in Chapter 6. The results are compared and analyzed relative to other data on polypropylene glycol. Where applicable, the data are compared to Mode Coupling Theory predictions. A summary of the results and suggestions for future efforts are contained in Chapter 7. References 1 Structure and Dynamics of Glasses and Glass Formers,Vol. 455, edited by C. A. Angell, K. L. Ngai, J. Kieffer, T. Egami, and G. U. Nienhaus (Materials Research Society, Pittsburgh, PA, 1997). 2 DisorderedMaterials and Interfaces, Vol. 407, edited by H. Z. Cummins, D. J. Durian, D. L. Johnson, and H. E. Stanley (Materials Research Society, Pittsburgh, PA, 1996). 3 Supercooled Liquids: Advances and Novel Applications, Vol. 676, edited by J. T. Fourkas, D. Kivelson, U. Mohanty, and K. A. Nelson (American Chemical Society, Washington, DC, 1997). 4 M. D. Ediger, C. A. Angell, and S. R. Nagel, J. Phys. Chem. 100, 13200 (1996). Chapter 2 Mode Coupling Theory 2.1 Mode Coupling Theory Background One of the major theories of the liquid-glass transition, Mode Coupling Theory (MCT) has received much theoretical and experimental attention in recent years. 1-3 Several major reviews of Mode Coupling Theory, its mathematical formulation, and the experimental evidence to support or contradict the theory are available in the literature. The review by Cummins et. al.4 summarizes the mathematical results and presents a wide variety of experimental evidence. The presentation of Gotze et. al.5,6 begins with a discussion of typical features in the relaxation of supercooled liquids. Subsequently, the theory is presented in a detailed mathematical treatment in response to these observations. Mode Coupling Theory and its predictions are summarized in this section for completeness and future reference in the discussion of results and comparison to literature. Mode Coupling Theory attempts to explain the glass transition as a kinetic phenomenon controlled by non-linear coupling between modes. Each mode, labeled by wavevector q, is described by an equation of motion for the density autocorrelation function, (I(t). (t) = (2.1.1) q0) (P S (2.1.2) (t) + QO,(t) + Mq(t -t') (t')dt' = 0 where Q is a characteristic microscopic oscillation frequency and S, is the static structure factor. The substance of MCT is contained in the memory function (2.1.3) Mq(t - t') = C [yq(t - t') +mq(t - t')] The memory function contains two terms, a regular damping term, and the relaxing term, m,. In the simplest version of the theory, modes are quadraticly coupled by a coupling constant in the relaxing term. (2.1.4) mq(t)= V(2)(q,ql,q2) q ,q2 qt (t) q2 (t ) The theory becomes richer, albeit less mathematically tractable, with increased complexity in the coupling constant. Without the relaxing term, or in the weak coupling limit where the coupling constants approach zero, (2.1.2) reverts to the standard harmonic oscillator equation of motion with frequency Qiq and damping rate y,, which describes the behavior of a simple, non-viscous liquid. (2.1.5) q,(t) + q qbq (t')+ q,(t) = 0 The MCT equations (2.1.2) now form a closed set of equations that can in principle be solved numerically. The coupling constant V(2 ) is a function of the static structure factor, and contains the temperature dependence in the MCT formulation. The temperature dependence of the coupling results in a slowing down of the dynamics underlying the glass transition. As the system is cooled, a critical value V(2 )c is reached at a critical temperature T,. At this point, the feedback between the modes is strong enough to cause structural arrest. A single schematic equation can be obtained by restricting the modes to modes at q = qo near the peak of the static structure factor, where the coupling is strongest. Each relaxation function q,has identical time dependence due to symmetry, so the single MCT equation is (2.1.6) (t) + n 7(t) + C20(t)+ IV(2)b 2 (t- (t) (t')dt' = 0 As with the case of the full set of equations, at the critical value V(2)c, reached at T., the structural relaxation slows down, resulting in an ergodic to non-ergodic transition at To. Experimentally testable predictions are made in the vicinity of this transition, often formulated in terms of a separation parameter (2.1.7) a = (T - T)/T,. Figure 2-1 displays 0(t) and 0((o) for both the idealized and the extended theory, discussed below. In a qualitative, physical picture of relaxation, MCT predicts a two step relaxation function in time, or two peaks in the frequency spectrum. There is a fast relaxation, called the P relaxation. The second feature is a slow relaxation, which is referred to as structural relaxation, or a relaxation. The a relation is linked to the coupling term and thus shows a dramatic temperature dependence. The a relaxation extends to infinite time, or zero frequency at Te, according to the discussion above. _ I I~ 010S 06. I'I nr. J0 020 Os o" • 04 01 0' 0' u' 0 0 106 4 I. 10 I Figure 2-1: Illustration of the relaxation function in time, and the relaxation spectrum in the frequency domain. The upper illustrations represent idealized Mode Coupling Theory, while the lower figures are derived from the extended theory. The figure is reproduced from Ref. 4 Approaching T , the relaxation time and the viscosity diverge to infinity, consistent with structural arrest. MCT predicts that this divergence follows a power law temperature dependence (2.1.8)c Ocr 17 IT- Tc-r Viscosity data at high temperatures have supported this power law behavior, but with a measured TC value significantly above the glass transition temperature T,. 4,7 This immediately challenges the theory, as the viscosity does not diverge to infinity, but has measurable values down to the glass transition temperature, where the viscosity reaches 1013 poise. In addition, the a relaxation, which is arrested at Tc in the idealized theory, is clearly observed in experiments at temperatures approaching the glass transition, below the To value determined by the viscosity prediction. To address this issue, Mode Coupling Theory has been extended to include thermally activated process which restore ergodicity below Tc. The viscosity and a relaxation continue to vary smoothly through Te, which is now referred to as a crossover temperature. This version of the theory is referred to as extended Mode Coupling Theory, as opposed to idealized Mode Coupling Theory. Many of the results of the idealized theory are not greatly modified by the extended theory, and most experimental tests have only considered the idealized version of the theory. The strength of MCT is that there are several experimentally testable predictions derived from the solution of the MCT equation. One is the power law temperature dependence of the viscosity, mentioned above (2.1.8). As the relaxation time is proportional to the viscosity, it is also expected to exhibit power-law temperature dependence. Many of the predictions of MCT relate to the time dependence of the relaxation function, or equivalently, the frequency dependence obtained by Fourier transformation. The relevant quantities are defined by (0o) = 0'(W) + iq"() (2.1.9a-c) '(mc) =j(t) cos(ot)dt 0 (t) sin(ot)dt 0"() = 0 The imaginary part of the susceptibility, X, is related to the imaginary part of the relaxation function, t, by a frequency factor (2.1.10) X"(Co) oc coo"(Co) The prediction made by MCT concerning T, which is experimentally observable at, above, and below TC describes an anomaly in the variation of the Debye-Waller factor with temperature. The Debye-Waller factor is defined as the strength of the ax relaxation. In the frequency domain, this corresponds to the integrated area of the quasielastic line, the relaxation spectrum. (2.1.11) fq= I f "(o))d I|m< ,. =c , d(lnw) In o<ln q where oc is a cutoff frequency, set to a value high enough to completely include the ac relaxation in the integration. In the time domain, (2.1.12) fq = q(t Y ) Again, the turn-on time is chosen before the beginning of the slow relaxation. As long as any elastic features are well separated from the relaxation in time or frequency, the exact value of the cutoff time or frequency should not affect the measure value of the DebyeWaller factor. It may seem contradictory to speak of the "strength" of the a relaxation below To, as in the idealized theory the a relaxation is arrested at Tc. In this context, the concept of ergodicity gives a more physically intuitive picture of the Debye-Waller factor. Given the initial condition 4(t=0) = 1, the Debye-Waller factor, or the non-ergodicity parameter, is a measure of how far the long time limit of (t) is from zero. The amplitude of this non-zero long time limit is a measure of the non-ergodicity of the system. The increased strength of the ca relaxation implies a corresponding decrease in the strength of the 3 relaxation. These concepts can be visualized with the aid of Figure 2-1. In the low wavevector limit, the Debye-Waller factor can also be determined from acoustic measurements. 8 (2.1.13) fo =-1 =- , M0 and M are the elastic modulus in the limit of zero frequency and infinite frequency respectively. co and c, are the zero and infinite speed of sound, respectively. Similar to the cutoff frequency, in practice, zero frequency means frequency much lower than the c relaxation frequency range, and infinite frequency is of the order of co, above in (2.1.11) and (2.1.12), i.e., above the a relaxation spectrum. Since the Debye-Waller factor is a wavevector-dependent quantity, f, and fo will differ in absolute magnitude, but the temperature dependence, regardless of the method of measurement, is predicted to be the same. The prediction derived from MCT for the temperature dependence of the Debye-Waller factor (2.1.13) is written in terms of the separation parameter o, and the critical amplitude hq. h (2.1.14)=f (,OTT) (o < 0, T > T ) fq(T)= fq At low temperatures near the glass transition, there is a smooth transition from the glassy behavior, which is similar to the temperature dependence of the Debye-Waller factor for a classical harmonic solid, Inf, oc -T. The strength of the relaxation decreases as yo1/2 as the material warms, and at the crossover Tc, the relaxation strength reaches a steady level. This prediction cannot be exactly tested in the formulation stated above. Making a measurement of zero for the Debye-Waller factor would require an experiment with infinite frequency or time resolution, in order to select the appropriate cutoff frequency. Taking into account practical experimental considerations, the prediction is restated in terms of the effective Debye-Waller factor. Sfq(T) = f + h f,,() = + () + (a) (a > 0,T < T) (o < o, T> T The linear temperature dependent term is weak and is rarely included in the experimental analysis. Other predictions made by MCT concern the susceptibility spectrum (2.1.10). Although such experiments are not included in this thesis, these are also theoretically testable through ISTS, by measuring the acoustic modulus spectrum, given a large enough wavevector range. These predictions are often tested by frequency domain light scattering. 4 As predicted by MCT, the susceptibility spectrum should show two features, a high frequency feature referred to as P relaxation, and a lower frequency peak corresponding to a relaxation. If the frequencies are well separated, a minimum is observed between the features. The frequency window where this minimum can be observed is a function of temperature. In practice, accurate characterization of the minimum requires measurements over a wide frequency range including both the ca and P relaxation features. Due to the temperature dependence of the a relaxation, at high temperatures the features overlap and cannot be distinguished. As the a relaxation slows down with decreasing temperature, the features are well-separated, but the slow relaxation eventually moves beyond the low-frequency limit of the experiment, and the minimum can no longer be analyzed. The shape of the minimum in the susceptibility spectrum is determined in MCT by the exponent parameter k, which is specific to the material, and independent of temperature and wavevector. The magnitude of k is indicative of the degree of cooperativity, as seen in the original formulation of the equation of motion.9 (2.1.16) (t) + yI(t) + ~j YI\/ J CIt \L' jI.T Y/\L J + 4n(t) F0 (t -t')(t')dt' - t') 0(t') dt = 0 The minimum is described by (2.1.17) "()= [b(wm +±a ( n n (a + b) The critical exponents a and b are related to the exponent parameter X and are restricted to the values 0 < a <1/2 and 0 < b (2.1.18) 1. The relation between the exponent parameters is A= F_2-a =F2(l+b) F(-2a) F(1+2b) where F is the gamma function. Both the frequency and amplitude of the susceptibility minimum vary as power laws relative to Tc. (2.1.19a-b) z",, oc(T- T) Co"c (T-T Te There are equivalent power law relationships for the decay of the autocorrelation function in the time domain. The time domain signature of the high and low frequency features in the frequency domain is a two step decay of the autocorrelation function. The crossover between the two decays occurs at a time t, which is a function of temperature. (2.1.20) t, = t o / 10 1 " to is a microscopic time constant, the inverse of the microscopic frequency Q0 , and is independent of temperature and wavevector. For times longer than to but shorter than the timescale of the cX relaxation time ta, the decay function follows the power law decay (2.1.21) ~,(t) = fqC - Bh,(t/ta) The time constant t, varies strongly with temperature. (2.1.21) describes a power law asymptotic region of the a relaxation. In general, however, the ca relaxation is often well described by a Kohlrausch-Williams-Watts stretched exponential function. q(t) oc e (2.1.22) A similar relationship is predicted for the [3 relaxation, at times longer than the microscopic time to but shorter than t,. (2.1.23) qq(t) = f, + hq (t/to) The above power law decay describes the decay from the initial value )(0) to the plateau valuefqc, which is the same as the Debye-Waller factor (2.1.14) In the above expressions, the only role temperature plays in the shape of the relaxation function is in the scaling time t,, which varies as a power law relative to Tc (2.1.20). The wavevector dependence is contained in the amplitudes fq and hq. The exponents a and b are independent of time, temperature, and wavevector. These properties imply that the susceptibility spectra, or the relaxation functions, obtained at different temperatures can be superimposed simply through scaling by the relaxation time. The shape of the spectrum or relaxation function, determined by the power law forms in (2.1.21) and (2.1.23), is the same regardless of the temperature, above T c. This also implies that the shape of the a relaxation function, determined stretched exponential parameter 3 (2.1.22) is also the same. by the KWW 3 should not change with temperature above Tc. This behavior is referred to as time-temperature superposition. The exponents in these equations are the same as those in the frequency domain, and the crossover time t, is analogous to the frequency at the minimum of the susceptibility spectrum. It should be reiterated that these predictions in the idealized theory are for temperatures above Tc, consistent with the idealized theory prediction of structural arrest of the ocrelaxation at T c. Another point that merits reiteration is the fact that Mode Coupling Theory is formulated in terms of the autocorrelation function for density fluctuations (2.1.1). From this, all of the above predictions for time decay of the autocorrelation function, or the properties of the susceptibility spectrum, are meant to apply to density fluctuations. A valid experimental test of MCT must involve a quantity that is strongly coupled to density fluctuations. Before discussing experimental tests of mode coupling theory, some quantitative comments on time scales and frequency ranges will be made. A starting point on the fast time scale and high frequency extreme are molecular vibrations. A typical molecular vibration occurs in the tens of terahertz range, with a vibrational period of tens of femtoseconds. These motions are beyond the range typical of experiments designed to test MCT, and are not relevant to MCT, which is a monatomic liquid theory. Intermolecular motions occur on a longer time scale of hundreds of femtoseconds, with frequencies of a few terahertz. These modes have been detected in time domain experiments, and appear as the boson peak and other high frequency features in frequency domain experiments. As these motions are also not included explicitly in MCT, no direct discussion of their properties is necessary, although this contribution to the spectrum may be considered in data analysis procedures. MCT predictions begin after the microscopic time to = Q-.. The 3 and ca relaxations in MCT occur at longer times and lower frequencies in the supercooled liquid regime. Several characteristic spectra are reproduced in the review by Cummins et. al.4 The a and P relaxation features first make their appearance at frequencies of a few terahertz at temperatures well above the glass transition. As the liquid is cooled, the P relaxation remains in the terahertz or hundreds of gigahertz range, with the 3 contribution to the decay of the autocorrelation function occurring within a few picoseconds. The a relaxation is dramatically temperature dependent. It has been measured from the high frequency limit where it becomes separated from the 3 feature, to very low frequencies less than one hertz near the glass transition. These low frequency measurements correspond to relaxation times on the order of many seconds. The wide range in frequency or time required to characterize the c and P relaxations throughout the entire supercooled temperature range is experimentally challenging. The only technique to date with this wide dynamic range is dielectric spectroscopy. Various neutron and light scattering spectroscopies characterize the high frequency, high temperature regions, with frequencies from THz down to MHz. ISTS in its current incarnation covers the time range from a few nanoseconds to milliseconds. Fortunately, this time range corresponds to the time scale of the ca relaxation in the temperature range where T, is expected. This range, however, is not nearly fast enough to observe the P relaxation, or the crossover between a and 1relaxation. 2.2 Experimental Methods of Testing Mode Coupling Theory Neutron scattering is one of the most common techniques used to study the liquid- glass transition. 10 Several methods are combined for a frequency range from 100 MHz to 10 THz. The accessible wavevectors are of the order of inverse angstroms, corresponding to interatomic length scales. In this wavevector and frequency regime, neutron scattering directly probes atomic and molecular motions. The measured quantity in neutron scattering is the structure factor, S(q,o), or its Fourier transform, S(q,t). 1I This quantity is the same as the correlation function discussed in MCT. Coherent neutron scattering measures the pair correlation function, while incoherent scattering measures the self correlation function. Neutron scattering is often used to test the MCT prediction for the anomaly in the Debye-Waller factor. The measured frequency range often does not extend to low enough frequencies to permit integration of the spectrum. Instead, the Debye-Waller factor is obtained by measuring the plateau height in the decay of the correlation function in time. 10 As the full relaxation is not observed, this can be a source of uncertainty in the measurement. In addition to measuring the Debye-Waller factor, the MCT predictions regarding the behavior of the correlation function are investigated by neutron scattering.1 2 The o relaxation is most commonly fit to a Kohlrausch-Williams-Watts (2.1.22) function, and the invariance of the Kohlrausch 3 exponent with temperature is verified. However, due to the limited time duration of the signal, this test can only be conducted for temperatures well above Te, and the shape of the relaxation at and below Tc cannot be investigated. Light scattering studies overlap with neutron scattering measurements in the THz regime, and extend into the MHz frequency range. The exact mechanism for depolarized light scattering at high frequencies is not completely understood,4 but qualitative agreement with neutron scattering at high frequencies is normal. It is often assumed that the light scattering active modes are strongly coupled to density fluctuations, but the light scattering spectrum may include contributions from molecular rotational motions. The most common measurements made with light scattering test the MCT predictions concerning the shape of the susceptibility minimum (2.1.17). The exponent parameters X, a, and b are determined, and the power law temperature dependences of the susceptibility minimum and the minimum frequency are tested. Dielectric spectroscopy has an impressive frequency range from less than 1 Hz up to hundreds of GHz. 13 The drawback of testing MCT using dielectric spectroscopy is that the modes that are active in dielectric spectroscopy are coupled to the dipole moment, not necessarily to density fluctuations. In fact, deviations between the dielectric and light scattering spectroscopies can be observed, particularly at high frequencies. However, given the broad frequency range, dielectric spectroscopy is particularly valuable for studying the a relaxation throughout the supercooled temperature range. The glass transition has been studied using other tools, such as viscosity measurements, NMR relaxation times, specific heat spectroscopy, and other techniques. These experiments will be discussed as pertinent results are cited. References 1 Structure and Dynamics of Glasses and Glass Formers, Vol. 455, edited by C. A. Angell, K. L. Ngai, J. Kieffer, T. Egami, and G. U. Nienhaus (Materials Research Society, Pittsburgh, PA, 1997). 2 DisorderedMaterials and Interfaces, Vol. 407, edited by H. Z. Cummins, D. J. Durian, D. L. Johnson, and H. E. Stanley (Materials Research Society, Pittsburgh, PA, 1996). 3 Supercooled Liquids: Advances and Novel Applications, Vol. 676, edited by J. T. Fourkas, D. Kivelson, U. Mohanty, and K. A. Nelson (American Chemical Society, Washington, DC, 1997). 4 H. Z. Cummins, G. Li, W. M. Du, and J. Hernandez, Physica A 204, 169 (1994). 5 W. Gotze and L. Sjogren, Reports on Progress in Physics 55, 241 (1992). 6 Liquids, Freezing, and the Glass Transition, Vol. 1, edited by J. P. Hansen, D. Levesque, and J. Zinn-Justin (Elsevier Science Publishing Company, Amsterdam, 1991). 7 P. Taborek, R. N. Kleiman, and D. J. Bishop, Phys. Rev. B. 34, 1835 (1986). 8 M. Fuchs, W. Gotze, and A. Latz, Chem. Phys. 149, 185 (1990). 9 E. Leutheusser, Phys. Rev. A 29, 2765 (1984). 10 W. Petry and J. Wuttke, Transport Theory in Statistical Physics 24, 1075 (1995). 11 N. H. March and M. P. Tosi, Atomic Dynamics in Liquids (Dover Publications, Inc., New York, 1976). 12 J. Wuttke, W. Petry, and S. Pouget, J. Chem. Phys. 105, 5177 (1996). 13 A. Pimenov, P. Lunkenheimer, and A. Loidl, Ferroelectrics176, 33 (1996). Chapter 3 Impulsive Stimulated Thermal Scattering 3.1 Overview and Theory Impulsive stimulated scattering is a time domain laser spectroscopy that can excite and probe a wide variety of material modes. 1 Impulsive Stimulated Thermal Scattering, as implied by the nomenclature, results from heating of the material. The mode or modes that are probed consist of the response of the material to the heating. Using ISTS, relaxation dynamics of supercooled liquids are observed in the nanosecond to millisecond time range. Two picosecond infrared laser pulses at wavelength ke cross at an angle e in the sample, impulsively heating the material in a grating pattern created at wavevector A 2 sin(2) Heating results from absorption into vibrational overtones. A probe beam diffracted at Bragg angle monitors the density evolution of the grating, as the index of refraction varies with the density. The signal shows the density response dynamics. The theory of ISS, and its relationship to frequency domain light scattering, has been developed in detail. 1 In the limit of ideal time and wavevector resolution, the ISS signal is directly proportional to the square of the Green's function (impulse response function) describing the material. In the case of ISTS on supercooled liquids, the material under consideration is a complex fluid in the hydrodynamic (q-0O) limit. The generalized equations of hydrodynamics are solved for the Greens function response of density to sudden heating, GpT(q,t) 2 . In the Debye approximation, the result is (3.1.2) G,,(q,t) = A[e -r ' -e-r It. cos(2xrco,t)] + B e -rH' -e 0R where FH is the thermal diffusion rate, FA is the acoustic attenuation rate, cA = wA/q describes the speed of sound and acoustic frequency, co is the zero frequency speed of sound, and tR is the relaxation time. The relaxation dynamics of supercooled liquids universally show a non-Debye relaxation, and are more commonly described by a Kohlrausch-Williams-Watts (KWW) stretched exponential function. Then the ISTS signal is (3.1.3) I(q,t) oc G,, (q,t)2 = {A[e-' - e" - cos(2rco ,t)] + B[e- '' _ r ')]}2 In principle, this expression can be generalized to include other relaxation contributions, but this form has been found to give good fits to the data at low frequencies without any additional terms. For the KWW form, the average relaxation time is given by (3.1.4) (r) = FR The ISTS signal contains three modes, acoustic, structural, and thermal, with respective amplitudes A, B, and A+B. These modes are analogous to the Brillouin, Mountain, and Rayleigh modes observed in frequency domain light scattering. The relaxation strength is called the Debye-Waller factor or the non-ergodicity parameter, and is given by the amplitude ratio 2 fCoo = I-- (3.1.5) c. B A+B This expression is valid when the acoustic, structural, and thermal modes satisfy the condition, A>>FR>>H. This measurement is equivalent to the integration of the Mountain mode in light scattering or the integration of the quasielastic peak in neutron scattering. This measurement is used to test the predictions of Mode Coupling Theory. 3.2 Experimental 3.2.1 Nd:YAG laser The excitation source for all ISTS measurements in this thesis is a Nd:YAG laser, originally manufactured by Spectron, and modified to produce the desired output. The detailed cavity design is described elsewhere. 3 The original continuous wave (cw) laser is modified by incorporating a mode-locker, q-switch, and Pockel's cell for cavity dumping. When operating optimally, the output consists of 120 ps pulses with 700 microjoules per pulse, at a variable repetition rate from tens of Hz to 1 kHz. Frequently, when the experiment doesn't demand short pulses, the cavity length is detuned for a pulse duration in the vicinity of 200 ps, achieving more stable operation. The mode-locking frequency is generated by a radio frequency source, which is amplified for the mode-locker. The driver for the Pockel's Cell, manufactured by Medox, includes a frequency divider. The mode-locking frequency is variably divided by the Medox controller to determine the repetition rate of the laser and produce a trigger for the Stanford digital delay generator. One output from the delay box triggers the q-switch. This same output is again variably delayed by the Medox controller to fire the Pockel's cell. The other outputs are available to control the probe laser timing, or serve as triggers for detection systems. Optimal operation of the laser requires careful attention to several diagnostics. Since the laser cavity contains multiple elements with several adjustments each, no one diagnostic is sufficient to optimize the laser for energy, stability, and pulse duration. Among the properties that must be monitored are q-switch profile, pulse duration, energy, and cw-steady-state level. Once the laser is running well, it will run for several days without serious adjustment. The temporal shape and stability of the q-switch envelope are the best diagnostics of overall laser energy and stability. The cavity alignment should be optimized so that the envelope rises as quickly and as high as possible. Care should be taken to iteratively optimize the cavity alignment and the alignment of the signal onto the photodiode. A large area photodiode should be used to eliminate this complication. The cw-steady-state level maintained between q-switch bursts should be set to achieve the desired balance between energy and stability. The higher this level, the more stable the output, but at the expense of output energy. Once a stable q-switch envelope is achieved, cavity-dumping should be optimized by Pockel's cell alignment and timing. The timing should be set to sharply cut off after the highest pulse in the q-switch envelope. The alignment of the crystal should be adjusted to minimize any trailing or leading pulse, as seen on a photodiode monitoring the laser output. Typically, the output easily saturates a photodiode, however, saturation by the peak pulse may be necessary to see the trailing or leading pulses. An extinction ratio of 500:1 can be achieved. Mode quality is also an indication of good Pockel's cell alignment. The lower the time constant controlling the timing, the better the laser alignment, as a low time constant is indicative of a quickly rising q-switch envelope. The pulse duration depends on the relationship between the mode-locking frequency and the cavity length, and is roughly determined from a scanning autocorrelator. pulsing". Autocorrelation is necessary to ensure that the laser is not "double- The Medox Pockel's cell currently in use was not originally designed for internal cavity operation. The crystal is parallel cut, and can create etalon effects in the cavity. This results in a "double" pulse in time, two 100 picosecond pulses separated by 100 ps. This behavior has been observed with both a streak camera and an autocorrelator. With a wedge cut crystal in the cavity, this behavior could not be duplicated. One "quick-and-dirty" method of estimating the pulse duration is to watch the stability of the cw level. Unfortunately, this is least stable when the pulse is shortest, or the laser is double pulsing. If a short pulse is not necessary, the cavity length can be optimized for stability. The laser has not been demonstrated to operate with pulses longer than 300 ps. Pulse duration has been found to be largely insensitive to minor adjustments in the cavity alignment. Major alignments, such as TFP angle, Pockel's cell position, Brewster angle for the mode locker and q-switch, and Bragg angle for the mode locker should only be adjusted in extreme situations. Under typical operating conditions, the only adjustments that should be made are to the end mirrors, Pockel's cell alignment, cavity length, and timing. 3.2.2 Generation of the excitation grating The "traditional" method of generating the diffraction grating in ISTS experiments is to manually align the excitation beams at the desired angle. The probe is also manually aligned to be incident on the grating at Bragg angle. Figure 3-1 illustrates the optical configuration. The excitation beam is split into two arms with a 50% beamsplitter. The two arms are aligned onto the sample, and focused in the direction perpendicular to the wavevector. Timing can be controller with the translation stage, but is not a sensitive adjustment using 100 ps excitation pulses. The probe is focused with an additional cylindrical lens parallel to the wavevector to produce a spot as close to round as possible. To generate small wavevectors, with angles less than 2 degrees, a combination of a dichroic beamsplitter and an additional 50% beamsplitter for the excitation may be used. Alignment is optimized by visual or electronic observation of the signal. The signal follows a long path from the sample to the detector, to facilitate separation of the signal from scattered light from the transmitted probe beam. Transmitted excitation light can be separated using a color filter. All data in this thesis were obtained using the method described above. An alternative to the "traditional" method of manually aligning the two arms of the excitation to cross at the desired angle in the sample is to use a phase mask. A phase mask is a piece of fused silica with a pattern etched to a specified depth. Figure 3-2 shows the design for a set of masks designed for several applications, including ISTS. The drawing contains the geometry of the design for the 4 inch darkfield chrome on quartz mask. The patterns consist of an array of vertically oriented bars and spaces with the widths of the bars equal to the spacing between the bars. The value in microns for the width and spacing of each pattern is given by the number in the box. An example for 3 microns is shown immediately at the bottom of the picture. When an optical beam is passed through the phase mask, multiple higher orders of diffraction are created. The diffraction angle is determined by the feature size of the mask and the wavelength of the light. The relative amplitude of the diffraction compared to the zero order is also determined by the wavelength of the light and the etch depth of the mask. For an etch depth equal to the light wavelength, virtually no light is transmitted in the zero order. _~~~ ~ _~ _~_ ___ Excitation Beam Probe Beam Cylindridal Lens V Translation Stage ST=50 % R=50% Translation Stage Translation Stage Cylindrical Lens Sample Diffracted Signal Figure 3-1: Illustration of the optical setup for generation and probing of the grating. Not drawn to scale. __ ~_ _ C_ ___ _ ___ 5 mm 4mmI[mm H 1mm 12 41231F ] F 61 m 4 or 10 mm as indicated H 2 mm 1 253011 1 1125]0 JpJ 12.5 13.5 14.51 5.5 16.51 8 5 mm 10 4mmI MM 21..5 H 12 14.5 5.51 6.51 37 35] 10 14 20 LII6122 11261 25 1 35 111 131 161 20 I 1 mm 12 jl14 S10 121 15 18 20 5mm 2 H ~~~~~~~~3h1~~~13 [H F9 3 43 6 o 84o 1130 1 mm 1. 10 mmI H1 5 2 50 100 H 1 mm I8F1E196 EWW SmmIE H 1mm 3.00 ll'''' H 3,00 Figure 3-2: Phase mask design showing patterns obtained for ISTS, ISRS, and other applications. 40 22 24 25 30 The +1 and -1 orders of diffraction are recombined with an imaging system to create a grating with fringe spacing determined by the phase mask feature size and the properties of the imaging systems. All wavelength components within the bandwidth of the optical pulse create a grating with the same wavevector. This has the advantage of eliminating the wavevector spread associated with the bandwidth of an ultrafast pulse. If excitation and probe beams pass through phase masks with the same feature size, the probe is automatically incident at Bragg angle regardless of the wavelength. This eliminates one major source of error in the alignment by the traditional method. Another advantage of the phase mask method of generating gratings is that since the probe is split in two, one arm can act as a "finder" beam for locating the diffracted signal in space. This is particularly useful if the signal is weak, or the probe wavelength is not visible by eye. Alternatively, the "other" probe beam can be attenuated and used for heterodyne detection. Figure 3-3 shows a simple imaging system using a phase mask. The mask is placed at a distance F from the first lens, the lenses are separated by the sum of their focal lengths, F + F2, and the sample is a distance F2 from the second lens. More complex imaging systems can be designed, but have not yet been applied to ISTS experiments. If F = F2, the imaging is 1:1, and the grating wavelength is equal to 2 times the spacing on the mask. Other imaging ratios can be selected to change the wavevector of the grating. 3.2.3 Probing the grating The probe source for the experiments in this thesis is a continuous-wave Argon laser (Lexel 3500) operated with an etalon for a single line (514.5 nm), single mode beam. The samples are transparent at this wavelength, reducing the possibility of signal created by probing effects. The output is electro-optically gated to produce a quasi-cw square pulse of arbitrary duration, to reduce heating of the sample. Careful alignment of the gate device will achieve a pulse of constant amplitude after some initial ringing. Very long pulses on the order of several milliseconds did not remain flat due to temperature changes in the gate device, so to measure very long time signal gating was not used. In this case, the sample was exposed to the beam for as limited a time as practical, particularly at low temperatures. reduce heating effects. The laser was run at the lowest possible power to Phase mask Lens with f=F1 Excitation beam Excitation beam Probe beam "Lens I l with f=F2 i VSample i Diffracted Signal Figure 3-3: Illustration of a simple imaging system for generation and probing of a grating using a phase mask. Recently, a diode laser is implemented as a probe. The diode used is an SDL 543 l-G1, at 830 nm with a maximum output of 200 mW. The beam is reasonably well collimated to a spotsize of 1 by 4 mm, using a collimation package from Thorlabs. A quasi-cw pulse is produced by triggering the diode laser power supply. The diode laser has the advantages of simple, reliable operation. Photodiode materials are often more sensitive at 830 nm than for visible wavelengths. Since the wavelength is similar to the 1064 nm excitation wavelength, one phase mask etched at either 800 or 1064 nm provides sufficient diffraction efficiency for both wavelengths. The major disadvantage of using the diode laser is that the signal is no longer visible by eye, even with an IR viewer and IR sensitive card. This difficulty is largely overcome by using the "finder" beam generated by the phase mask. 3.2.4 Detecting the Signal The signal is detected electronically with a fast photodiode and a digitizing oscilloscope. A Tektronix Digital Signal Analyzer (DSA) 602 with 11A72 1 GHz bandwidth plug-in is used for all data acquisition function. One of two fast amplified photodiodes was used to detect the signal. A photodiode manufactured by Antel (ARXSA) has a bandwidth from hundreds of MHz to dc, with a gain of 35000 Volts/Watt. Alternatively, a photodiode from Hamamatsu, with bandwidth from 2 GHz to 3 MHz and gain of 250,000 Volts/Watt is used. The high gain of the Hamamatsu diode makes it very easy to detect the signal, but the low frequency roll-off prevents measuring any long time dynamics. The use of these two diodes, and the procedures used to merge the data, will be discussed further in the section on Data Analysis. 3.2.5 Temperature Control All of the samples for the experiments in this thesis were contained in teflon- coated aluminum sample cells. The windows of the cell are held in place with a flange and o-ring, to permit some contraction of the sample volume as the material cools. The pathlength of the cell is approximately one inch, so the depth of the optical grating is always determined by the angle and the excitation spot size. A resistance thermal detector is sealed in the cell, immersed directly in the liquid. Using this design, the sample maintains good optical quality throughout the supercooled temperature range, but usually develops cracks around the glass transition temperature. The sample is mounted on a cold finger in a cryostat. Either a closed cycle helium refrigerator, or a liquid nitrogen cold finger were used for these experiments. The sample is directly attached to a copper block at the base of the cold finger, using indium or copper mixed in vacuum grease to achieve good thermal contact. The copper block contains resistive heaters for temperature control. A platinum resistor monitors the temperature of the copper block. A Lakeshore temperature controller is used to monitor the temperatures of the sample and the copper block. The temperature data are fed to a computer through a GPIB interface. 3.2.6 Automation The ISTS experiment on supercooled liquids is particularly well suited for automation. Once the signal is acquired and optimized, little manual adjustment of the laser or optics is required. A typical temperature range at a single wavevector for a supercooled liquid will span 150 to 200 degrees, with temperature increments of 5 K or 1 K depending on the conditions. It takes at least 10 minutes per temperature for the sample to reach thermal equilibrium, resulting in an experiment that takes 8 or more hours, the majority of which is spent equilibrating the temperature. A Labview® code has been written to automate the process of changing the temperature, determining the equilibrium temperature, and acquiring data with the oscilloscope. Labview is a graphical programming language designed to interface with laboratory equipment. The temperature controller and oscilloscope both have a GPIB interface. Briefly, the user creates an input file with the temperatures for data acquisition. The user also initializes and saves the acquisition parameters for the oscilloscope. Then the program takes over the experiment. The temperature is set, and the oscilloscope begins averaging the data once thermal equilibrium is reached. Temperature readings are also taken during the 20 to 30 seconds it takes for the oscilloscope to complete averaging to ensure that the sample temperature is not drifting or cycling. Although data analysis has not yet been incorporated into the routine, it is also possible to create an interface between Matlab® and Labview, making it possible to automate a rough fit of the data while the sample is equilibrating at the next temperature. Phase masks create even more possibilities for automation. The mask could be translated automatically to change from one wavevector to another. Given careful imaging of the signal onto the detector, no manual adjustments would be required. This would be particularly valuable if an experiment required measurements at multiple wavevectors at exactly the same temperature. 3.3 Heterodyne Signal When a heterodyne signal is deliberately desired and controlled, it can be used as a tool for amplifying a weak signal, or favorably changing the time dependence of the signal. However, when there is scattered light, there can be uncontrolled heterodyning between the scattered light and the signal. Any contribution from a heterodyne signal will have an effect on the parameters determined from the fit to the ISTS signal. The fitting routine assumes that the signal is purely (3.3.1) S(t) oc GP,(t)l However, if there is significant scattered light, the signal with a cw probe may become (3.3.2) S(t) oc c lGp,,(t ) + c2 Gp,,(t)l2 with experimentally uncontrolled amplitudes. A third component, unlisted in (3.3.2) is a dc signal which cannot be distinguished from scattered light. This contribution does not affect the analysis, and only affects the experiment by possibly saturating the detector. A signal with c, = c2 = 0.5 has been simulated using values of the parameters typical of glycerol at low temperature and wavevector. The modes are clearly separated in time, permitting measurement of the structural relaxation time and the Debye-Waller factor. As c2 is of the order of magnitude of the scattered light, and c, is very small due to the small diffraction efficiency, the amplitudes selected above are not unreasonable. Theoretically, c, should average to zero because there is no fixed phase relationship between the signal and the scattered light. It is assumed here that there is at least some phase preference permitting observation of the heterodyne signal. The lack of control over the phase of the light prohibits conducting the experiment to observe the heterodyne term overwhelmingly. The original values, and the fit results are listed in Table 3-1. Table 3-1: Simulated and Fit parameters for ISTS signal with a mixture of homodyne and heterodyne signal. Errors in the fit results are +/-5% or less A Original Value 0.4 Fit Result 0.56 Deviation - FA 16 27 +70% _) A FH B FR P t DWF 50 50 0% 0.0001 0.77 0.02 0.60 0.03 0.66 0.000067 0.61 0.027 0.59 0.04 0.52 -33% +35% -2% +33% -21% Due to geometric considerations, particularly the proximity of the signal to the transmitted probe, heterodyning is more likely at low wavevectors. Heterodyne signal is also more likely to have a significant contribution at lower temperatures, as the decreased compressibility approaching the glassy state is accompanied by a decrease in diffraction efficiency. The scattered light level also increases at low temperatures. One signature of a heterodyne component to the signal would be an unusually small value of the thermal diffusivity. However, this measurement depends on the accuracy in the determination of the wavevector. If the extent of heterodyning varies within a wavevector, for example with translation of the sample, unusual variations in the rate of thermal diffusion would be observed. The acoustic frequency is unaffected, and the perceived damping is high. Once again, a large value of FA/q 2 could be an indicator, but is subject to error in the wavevector and in the damping rate. Determination of damping rates at low temperatures and low wavevectors is also prone to walk-off errors. The relaxation time is lengthened, but the value of 3 is relatively unaffected. As the relaxation time should be independent of wavevector, these measurements can be examined for evidence of heterodyne contributions. The large discrepancy in the DebyeWaller factor is of serious concern. amplitude factors. This change is the natural result of additional 1.0 0.8 cl=0; c2=1 ............. cl=0.5; c2=0. 5 0.66-60.4 0.2 S 0.0 .* I I 0 1000 ' I I ' 2000 ' 3000 I 4000 ' I 5000 0.8 1.0 0.6 0.4 .... 0.2 0.0 1E-3 0.01 0.1 1 10 100 1000 10000 Time (pts) Figure 3-4: Comparison of simulated data for ISTS signal with different amplitudes of homodyne and heterodyne contribution. The data are simulated according to Equation (3.3.2). 100 000 Simulated Data and Fit cl = 0.5; c2 = 0.5 1.0- 0.8- 0.6- 0.4- 0.2- 0.0 - I _ _ 1000 1000 I 2000 3000 4000 I000 5000 log Time (ps) Figure 3-5: Simulated data with a heterodyne component (3.3.2) and fit to homodyne ISTS signal (3.1.3). Data are presented on a linear scale, with the inset showing the acoustic signal. The simulation and fit parameters are presented in Table 3-1. 1 Simulated Data and Fit cl = 0.5; c2 = 0.5 1.0 0.8 0.6 0.4 0.2 0.0 *1...................................I_______ ________~~ 1E-3 0.01 0.1 1 10 100 1000 _____ 10000 100000 log Time (ps) Figure 3-6: Simulated data with a heterodyne component (3.3.2) and fit to homodyne ISTS signal (3.1.3). Data are presented on a log scale, with the inset showing the acoustic signal. The simulation and fit parameters are presented in Table 3-1. 3.4 Data Analysis Methodology 3.4.1 Accounting for detector response The derived function for the ISTS signal assumes ideal time resolution. In practice, this is not the case, and the observed signal is affected by the excitation pulse duration, the probe pulse profile, and the response function of the experimental detection system. Specifically, the observed signal is given by 4 (3.4.1)S,,h,,(t) oc fR(t - t') Acw O(t - t')G,(t' -t") A, (t")dt" dFt' For all of the experiments in this thesis, the limiting factor in the time resolution is the finite response time of the electronic detection system, which is much longer than the pump pulse duration. When using a quasi-cw probe pulse profile and a detector with dc response, the convolution over the pump has no effect on the signal. The response function of the detection system is determined by measuring the response of the system to a pulse much shorter than the rise and fall times of the system. The detection system includes the digitizing oscilloscope, amplified photodetector, and any connectors and cables. This measurement was made for both a sub-picosecond pulse and the excitation pulse for the ISTS experiments. The response to the picosecond pulse was slightly longer in duration with less ringing, and this response was used in the data analysis. The Antel amplified photodiode has a frequency response from dc to several hundred MHz. The fall-off of the response starting above 200 MHz strongly affects the acoustic ISTS data. The impulse response must be deconvolved from the data to obtain a more accurate representation of the material response. The Hamamatsu detector has a level high frequency response, but a low frequency roll-off at 3 MHz. In both cases, the oscilloscope digitizes the signal with a bandwidth of 1 GHz. differences in the appearance of Figure 3-7 shows the identical data from glycerol, taken with the two detectors. Comparsion of ISTS Data from Antel and Hamamatsu Detectors Antel 0 100 200 30 0 ) 400 50C Hamamatsu -100 0 100 200 300 400 Time (ns) Figure 3-7: Example of raw ISTS data taken with two different detectors. The Antel data show attenuation of the high frequency component of the signal. The Hamamatsu data show a low frequency roll-off in the long time signal. 500 Accounting for detector response may not be necessary if the object of the experiment is simply to measure the parameters such as acoustic frequency and thermal decay rate. However, for ISTS experiments, one of the values determined is the ratio of high frequency to low frequency response. In this case, even though a high frequency oscillation may easily be observed and measured, the amplitude of the high frequency response relative to dc can be significantly affected by the bandwidth characteristics of the detection system. Deconvolution of the detection response is also important in making very accurate measurements of acoustic frequency and especially the damping rate near the upper frequency limit of the detection system. In the frequency domain, the Brillouin peak, represented by a Loretzian lineshape, is multiplied by a decreasing frequency response. When fitting the data, this can result in a systematic underestimation of the frequency. The damping will also be affected since the lineshape is no longer strictly Lorentzian. The convolution described in Equation (3.4.1) can be rewritten as a product in the frequency domain. (3.4.2) S0I,,() c [R(co) x G,.(wc)] Z One option for deconvolving the impulse response is to transform into the frequency domain, and do the division in Equation (3.4.2) to obtain GT(co). The frequency domain response R(co) of the system is given by the Fourier Transform of the temporal impulse response. However, practical considerations limit the accurate measurement of the relative amplitudes R(o) at high and low frequencies determined solely from the fast temporal response. At the low frequency end, the data must continue indefinitely in time to determine the dc amplitude. To measure the high frequency spectrum, the data points must be closely spaced. The duration of the impulse response is on the order of a few nanoseconds. Quantitatively, the sampling rate must be on the order of 1 point per 50 picoseconds to accurately measure the impulse response. For the Fourier transform to include frequencies as low as a few kilohertz, the total duration of the signal must be on the order of 1 millisecond. A 1 millisecond duration signal, sampled at a rate of 100 samples per second, has 107 points. The oscilloscope cannot acquire an arbitrarily large number of points with an arbitrarily long time duration for the data. Acquiring a shorter waveform and padding with zeros can give the appropriate properties of the data, but the number of points is computationally unwieldy. To take the frequency domain approach, the data must also be Fourier transformed. The ISTS data at short time commonly consist of 1024 or 2048 points at a sampling rate of 2 Gs/sec. In order to perform the discrete division with the response function, this data must have the same time axis as the response function. The data must be interpolated to the sampling rate of the impulse response, and extended to the same long time limit, on the order of 1 millisecond. Another potential pitfall is that if any ringing or aliasing occurs in the Fourier transform, the high frequency components will be amplified by the division of the response function which approaches zero at very high frequencies. The data after the division must subsequently be passed through a low bandpass filter designed to not recreate the original problem of high frequency attenuation. Finally, the data should be reduced to the original sampling rate so they can be easily fit and manipulated. The approach selected is to correct the data by deconvolution in the time domain. Matlab was used for its flexibility in data manipulation, its graphical capabilities, and its built-in convolution functions. Several key Matlab codes are included in the Appendix of this thesis. The procedure involves resampling the short time, acoustic part of the ISTS data up to the sampling rate of the real data for the impulse response. The measured impulse response is deconvolved from the resampled data. The data are subsequently resampled back down to the original rate. Without resampling, the ISTS data only have a few points in the duration of the impulse response. To maintain the low sampling rate, the impulse response would only be represented by 2 or 3 points. This method is prone to error in the measurement of the impulse response. In addition, the impulse response (Figure 3-8) cannot be fit to a functional form with a high degree of accuracy. Figure 3-9 shows the results of deconvolution of ISTS data from the Antel detector, at an acoustic frequency near 300 MHz. Note that the deconvolution doesn't seem to completely recover simple liquid response, with complete modulation of the first acoustic period. The reason for this is not completely understood. Perhaps the measured impulse response does not give sufficient quantitative information on the relative amplitudes of dc and high frequency response. As the data from the Hamamatsu detector show complete modulation, it is assumed that this is still an electronic, not a physical, effect. Antel Response to Picosecond Pulse 0.180.16 0.14 0.12 0.10 0.08 0.06 0.04 0.02 0.00 -1.6 -0.8 0.0 0.8 1.6 2.4 3.2 4.0 Time (ns) Figure 3-8: Measured Impulse Response of the Antel detector. This is the response used in the deconvolution of the ISTS data. 4.8 Comparsion of Original and Deconvolved Data Antel Detector Original " -100 0 100 200 300 400 _-- -100 0 100 200 300 500 Deconvolved 400 500 Time (ns) Figure 3-9: ISTS data taken with the Antel detector. The deconvolved data show an increased depth of modulation in the acoustic signal. The above discussion focuses on attenuation of the acoustic response in ISTS due to finite bandwidth of the detection system. In addition, the opposite problem on the low frequency end is encountered when using the Hamamatsu detector with a low-frequency roll-off in the amplifier. Figure 3-10 shows the response of the Hamamatsu detector to a quasi-cw pulse. The detector sees the rise and fall of the pulse, but doesn't give the dc signal. Response of Hamamatsu Detector to 200 [ts quasi-cw pulse I0 0 -50 0 / /I 50 200 250 300 Time (jis) Figure 3-10: Response of the Hamamatsu Detector to a quasi-cw probe, 200 [ts in duration. At present, no rigorous correction is made for this problem. Instead, the data are treated as if the probe pulse has an intensity that decreases at long times. The slope of the decay is estimated from the response of the Hamamatsu detector to the flat quasi-cw probe (Figure 3-10). The data are divided by this decaying function. Data from the Hamamatsu detector are only used until the end of the acoustic signal, after which the data from the Antel detector are used. The correction to the high frequency data improves accuracy in the combination of the data from the two detectors, but does not necessarily correct for different amplitudes of the acoustic and slow components in the Hamamatsu data. Comparsion of Original and Corrected Data Hamamatsu Detector Original _- 0 0 100 300 200 400 - 500 CCorrected 1 -100 0 100 200 300 400 Time (ns) Figure 3-11: ISTS data taken with the Hamamatsu detector. The corrected data show a long time signal more compatible with the data from the Antel detector in Figure 3-9. 500 3.4.2 Time Domain fits to the data The time domain data are fit to the signal derived from hydrodynamics theory. A Levenberg-Marquat algorithm 5 is used to conduct a non-linear least squares fit to the data. Whenever possible, the data are fit with equal weights to all points. When experimental conditions such as a poor baseline warrant, the higher quality portions of the data are weighted more heavily. 3.4.3 Alternate analysis of relaxation mode An alternative method to study the relaxation dynamics and measure the Debye- Waller factor would be to analyze the data using a method similar to one employed for neutron scattering measurements of the autocorrelation function. 6 Take the square root of the data to obtain (3.4.3) S' oc (A + B)e - "' - Ae-r'"cos(2roAt) - Be - (r The data at long time can be fit to a single exponential decay to obtain the quantities (A+B) and FH. The data are subtracted from the fit of the thermal decay, resulting in (3.4.4) - (r S" oc Ae-r" cos(2coAt)+ Be ') ' At times longer than a few times the acoustic damping FA, the signal is simply the decay function (3.4.5) S"'= b(t)= Bexp(-( 1 Pt) ) The relaxation function, in the form (3.4.6) - In = Fft -ln B can be fit to a power law equation to determine the parameters B, FR, and p. This method of analyzing the relaxation function should be far less dependent on the response of the detector at high frequencies, since the acoustic mode is not included in the analysis. The relaxation function (3.4.5) can be tested for the prediction of time-temperature superposition. The relaxation can also be tested for deviation from KWW behavior. References 1 Y.-x. Yan and K. A. Nelson, J. Chem. Phys. 87, 6240 (1987). 2 Y. Yang and K. Nelson, J. Chem. Phys. 103, 7722 (1995). 3 S. Silence, Ph.D. Thesis, Massachusetts Institute of Technology, 1991. 4 A. Duggal, Ph.D. Thesis, Massachusetts Institute of Technology, 1992. 5 W. H. Press, W. T. Vetterling, S. A. Teukolsky, and B. P. Flannery, Numerical Recipes in C (Cambridge University Press, Cambridge, 1992). 6 J. Wuttke, W. Petry, and S. Pouget, J. Chem. Phys. 105, 5177 (1996). Chapter 4 ISTS of fragile glass-formers Previous ISTS experiments have focused on fragile glass-forming liquids, specifically phenol salicylate (salol), 1 Ca 4K6(NO3) 14 (CKN), 2 and n-butylbenzene. 3 Salol and CKN have been several of the most studied supercooled liquids. In general, Mode Coupling Theory has had good success in describing the dynamics of these systems. Several figures are reproduced in this chapter for illustration of experimental tests of Mode Coupling Theory, and for future reference. As discussed in the previous Chapters, ISTS measures the a relaxation in the time domain, in the temperature range surrounding common values of Tc found in fragile glass forming liquids. The a relaxation has been observed in salol from 280 to 220 K (Tg = 215). Notice that since salol is a fragile liquid with relatively short relaxation times until the temperature is quite near the glass transition, the relaxation time can be measured to fairly close to the glass transition (Figure 4-1). In salol, the change in slope of the Debye-Waller factor is clearly observed from ISTS measurements (Figure 4-2). The value of To = 266 K is within experimental error of values obtained from neutron scattering. It is also interesting to note that a change in the behavior of P occurs in the temperature vicinity of Tc (Figure 4-3). This is not predicted by Mode Coupling Theory, as the a relaxation is theoretically arrested at Tc. However, this observation is in agreement with the physical picture of a change in the a relaxation dynamics at Tc. These data also serve as evidence that time-temperature superposition does not hold below Tc. Similar results are obtained from measurements on CKN (Figures 4-4 and 4-5). In this case, the "cusp" is more rounded than in salol. This effect is attributed to thermally activated processes, which are included in the extended version of Mode Coupling Theory. temperatures near T. As with salol, a change in the behavior of P is observed at A crossover temperature is also observed in butylbenzene, as shown by the Debye-Waller factor data in Figure 4-6. The 1 value also shows a decrease in the vicinity of Tc.4 Average Relaxation Time in Salol 10° 100 10 - 10-2 10-3 10-4 10 -5 - 1010 -6 -7 10 10 - 10o 200 220 240 260 280 300 320 340 360 380 400 T (K) Figure 4-1: Temperature Dependence of the Relaxation Time in Salol. Includes a comparison to Dynamic Light Scattering Data Debye-Waller Factor of Salol 0.3 L 230 240 250 260 270 280 T (K) Figure 4-2: Temperature Dependence of the Debye-Waller factor in Salol, in the q-*0 limit. A value of Tc = 266 K is obtained from a fit to the Mode Coupling Theory prediction. Stretching Parameter P vs Temperature 1.0 0.9 S0 0 Og 0.8 0.7 o q=B m-' 0.6 IT 11- 0.5 a q=B m-I V q=B m-1 * q=1.260 ptm * DLS - M PCS 0.4 1 200 [ I 220 I I 240 I I I 260 I 280 I 300 350 T (K) Figure 4-3: Temperature Dependence 3 in Salol. Data obtained from Dynamic Light Scattering is included for comparison. A change in behavior is observed near Tc = 266 K obtained from Debye-Waller Factor measurement. 400 Debye Waller Factor for CKN 0.66 0.64 0.62 0.60 A 0.0 0.58 A 0.56 I 360 I I I I 400 380 T (K) Figure 4-4: Temperature Dependence of the Debye-Waller factor in CKN, in the q->0 limit. A value of T c = 378 K is obtained from a fit to the Mode Coupling Theory prediction. A I 420 Stretching Parameter P of CKN 0.8 0.7 - 0.6 X++ 0.5 M v ~ WI----- O q=0.232 im A q=0.235 Vm A-M 0.4 - q=0.227 Vm - AC q=0.336 lm O q=0 423 im' T _ 0.3 X q=0.623 jlm - + q=0.896lm --- NS A PCSI 0.2 * PCS2 I 340 I 360 I 380 , I 400 420 * 440 DLS , I , 460 T (K) Figure 4-5: Temperature Dependence P in CKN. Data obtained from other experimental techniques is included for comparison. A change in behavior is observed near Tc = 377 K obtained from Debye-Waller Factor measurement. Debye-Waller Factor in Butylbenzene 0.54 0.520.500.480.460-" 0.44U 0.420.400.38- 140 U U 142 144 146 148 150 152 154 156 Temperature Figure 4-6: Temperature Dependence of the Debye-Waller factor in butylbenzene. The figure is reproduced from Ref. 3. 158 References 1 Y. Yang and K. Nelson, J. Chem. Phys. 103, 7732 (1995). 2 Y. Yang and K. A. Nelson, J Chem. Phys. 104, 5429 (1996). 3 L. Muller, Ph.D. Thesis, University of Texas, 1995. 4 DisorderedMaterials and Interfaces, Vol. 407, edited by H. Z. Cummins, D. J. Durian, D. L. Johnson, and H. E. Stanley (Materials Research Society, Pittsburgh, PA, 1996). Chapter 5 Impulsive Stimulated Thermal Scattering Study of Glycerol 5.1 Experimental background Glycerol has been extensively studied experimentally, in an attempt to extend Mode Coupling theory analysis to a glass-forming liquid of intermediate strength. Experimental techniques used to measure the properties of glycerol include neutron scattering, 1- 8 light scattering, 2,9, 10 dielectric spectroscopy, 11-14 non-resonant burning,15 specific heat spectroscopy,1 6 NMR8,17 and other techniques.18 hole The experiments most pertinent to testing MCT are summarized here, along with their results. R~ssler et. al.9 conducted frequency domain light scattering experiments to measure the low frequency Raman spectrum, in the frequency range from several hundred GHz to 10 THz. In this frequency range, a minimum in the susceptibility spectrum is observed for the temperature range from 298 K to 403 K. To analyze the data in the framework of MCT, the susceptibility spectrum, and the data at different temperatures are scaled according to z"min. The resulting data are fit to the MCT prediction for the shape of the minimum, with parameters b = 0.71 and a = 0.342. If the MCT restriction for the exponents is lifted, a better fit, particularly at higher temperatures, is obtained with b = 0.37 and a = 1.16. This violates the MCT restrictions for the magnitude of a and the relationship between a and b. Agreement of the data with the scaling relations for Z"mn and o,,i,, can be found for a = 1.16 from the unrestricted fit. A value of Tc - 300 K is obtained from the power law fits. If the parameters from the restricted fit are used for the power law fit to viscosity measurements, Tc = 320 K is obtained. The best fit to the viscosity data results in y = 2.5 and T, = 310 K. From the large value of a = 1.16 obtained in the unrestricted fits to the susceptibility, it becomes evident that the strong vibrational contribution to the spectra in the THz regime complicates the MCT analysis. To obtain a more realistic analysis, the vibrational component of the spectrum, obtained from low temperature data, is subtracted from the spectrum. The resulting master fit with the MCT restriction results in the same parameters a = 0.342 and b = 0.71. The best unrestricted fit is much closer to the MCT restriction, with b = 0.49 and a = 0.69. The resulting power law temperature dependence is insensitive to the change in exponents. The TC value of 310 K is significantly farther above the glass transition than values obtained for more fragile glass-forming liquids. For glycerol, Tg = 185 K and Tc = 310 K yields Tc 1.7 Tg, where typical values for fragile glass-formers are T, - 1.2 T,. The MCT predictions for the high frequency side of the minimum are tested directly and independently of the rest of the minimum, using the data minus the ca relaxation spectrum. The resulting data show a crossover from a power law with a - 1 to a = 0.34 closer to the frequency minimum. The conclusion is that MCT is only valid up to the frequency where the boson peak or the vibrational contribution complicates the high frequency spectrum. Simultaneous studies of neutron and light scattering are conducted to compare the two techniques and test the predictions of MCT. 2 The data from both techniques coincide at high frequencies greater than 1.4 THz, where the spectra are constant with temperature both above and below the glass transition. A free fit to the susceptibility minimum, discounting the MCT restriction, gives an exponent of a = 1, corresponding to the boson peak rather than the P relaxation. Data at lower temperatures show the power law low frequency wing of the 3 feature, and fits to this data for a more limited temperature and frequency range result in the parameters a = 0.54 and b = 0.43 without the MCT restriction and a = 0.32 and b = 0.61 with the MCT restriction. The latter fit poorly represents the minimum frequency. relaxation only results in b = 0.54 and k = 0.76. A fit to the c A power law fit to the minimum frequency using the a value obtained from the restricted MCT fit yields T, ; 225 K. However, this temperature cannot be used to fit the X"min data. A power law fit to the viscosity indicates a Tc value greater than 300 K. The light scattering data were more extensively analyzed in the framework of MCT. 19 An additional term for linear coupling is incorporated in the memory function (5.1.1) m(t) = v(t) +v 2 (t) 2 The coupling constants are allowed to vary with temperature, and the resulting calculated spectra are similar to the experimental spectra from glycerol. The exponent parameters, however, are not in agreement with the original MCT prediction. The value of To used in the calculations is between 223 and 233 K. No more specific prediction of Tc is made. The exponent parameters obtained are a = 0.314, b = 0.591, and X = 0.730. This fit includes the crossover from P relaxation dynamics to vibrational dynamics, as well as the a relaxation to p relaxation crossover. Glycerol has also been studied by dielectric spectroscopy.12,13, 20 There is a pronounced difference between the dielectric loss E"and the light or neutron scattering susceptibility at high frequencies in the THz regime. 13 The explanation offered is that the vibrational motions that contribute to the high frequency feature in the light scattering spectra are not coupled strongly to dipolar reorientations, which are probed in dielectric spectroscopy. The spectra below the minimum are in good agreement. From this agreement, it is concluded that "In the a-relaxation regime the reorientation of the glycerol molecules involves the tear and repair of hydrogen bonds which leads to the observed strong coupling of structural and dipolar relaxation".13,16 The susceptibility minimum is fit to the MCT prediction with exponenets a = 0.325 and b = 0.63, although the frequency range does not extend very far above the minimum. The power law relations are tested for the frequency of the minimum (Vmin), the magnitude of the minimum (E"mi), and the maximum of the a relaxation (vmx). Consistent fits are obtained with TC = 262 K. Another dielectric study yielded Tc = 248.8 K from a power law fit to the peak frequency. 21 Table 5-1: Experimentally determined exponents for the susceptibility spectrum of glycerol. Reference Method Exponents Tc (K) Wuttke et al. 2 Wuttke et al. Rissler et. a19 Rissler et. al Franosch et al. 19 Light Light Light Light Light 225 Schonhals, et. al. 2 1 Lunkenheimer et al. 13 Dielectric Spectroscopy Dielectric Spectroscopy a = 0.32; b = 0.61 a= 1 a = 0.342; b = 0.71 a = 1.16 a = 0.314; b = 0.591; k = 0.730 y= 3.65 Scattering Scattering Scattering Scattering Scattering a = 0.325; b = 0.63; k = 0.705 320 300 223-233 MCT restricted yes no yes no yes 248.8 yes 262 yes Glycerol has also been studied by the more unusual method of specific heat spectroscopy. 16 Specific heat spectroscopy measures the real and imaginary parts of the product of the constant pressure specific heat and the compressibility, Cp. IKdoes not vary with either temperature or the frequency in the ranges studied in this experiment, so the temperature and frequency dependences observed are attributed to the properties of cp. The real part shows a change in behavior from liquid-like to glass-like, with a lower cp in the glass. The transformation occurs at lower temperatures when the measurement is made at lower frequencies. This behavior is analogous to the change in the speed of sound with temperature observed in ISTS. The imaginary part of the heat capacity is treated as a susceptibility spectrum. Fits are generated to the Fourier transform of a KWW function, with P = 0.65±0.03 for all temperatures in the range from 200 to 219 K. The peak frequency from the spectrum is fit to the power law equation (5.1.2) v= vo[(T-)/ 0 ]a with parameters To = 169 K and a = 15.0. This equation is mathematically equivalent to the prediction of Mode Coupling Theory for the temperature dependence of the peak frequency of the a relaxation. However, the To value obtained is below the glass transition. In fact, the data are below the TC = 1.2 Tg seen for many fragile glass forming liquids. This result should not be discounted however, as MCT makes no statement as to the location of T. relative to the glass transition temperature. The acoustic properties of glycerol have been studied by an ultrasonic experiment. 22 The sound velocity and attenuation are measured as a function of temperature at 2 MHz, 10 MHz, and 27 MHz, very close to the frequencies associated with lowest wavevector ISTS measurements. The observed behavior shows the expected frequency dependent change from liquid to glass behavior in the speed of sound. There is also a temperature and frequency dependent maximum in the acoustic attenuation rate corresponding to temporal overlap of the acoustic period with the relaxation time. The longitudinal compliance is calculated from the acoustic parameters. The resulting susceptibility spectrum is fit to a KWW function with P = 0.60±0.05. The maximum frequencies from these spectra are combined with the maximum frequencies from the specific heat data, and again fit to Equation (5.1.2). The fit parameters are To = 175.5 K and a = 12.5. The same data are fit to a VTF equation with B = 2310 and TTF = 129. From the variety in the results described above, it is clear that there is a need for additional testing of MCT predictions for glycerol. ISTS is used to study the density dynamics specifically. So far, evidence of To obtained from the behavior of the DebyeWaller factor has not been reported, 23 and this study seeks that evidence. 5.2 Experimental Glycerol (99.5+%, <0.1% water, under nitrogen, Aldrich) was used without further drying or purification. The sample was transferred under nitrogen into a specially designed Teflon-coated cell that incorporates moveable windows to reduce sample stress and cracking. The sample was mounted onto a cold finger outfitted with resistive heaters. A resistance thermal detector immersed in the sample measures the temperature. Data were taken at temperature intervals of 5 K in the high and low temperature regions, and at 1 K intervals in the region where structural relaxation is well separated from the acoustic and thermal modes. For all temperatures, the sample equilibrated until the standard deviation of 10 consecutive resistance measurements was less than 0.005K. The sample maintained good optical quality over the entire temperature range from the high temperature liquid to the glass transition. Cracking of the sample prevents measurement at temperatures below the glass transition. The ISTS experimental design is described in detail in Chapter 3. The excitation pulses are 200 ps, 500 VLJ at X=1.064 Lm, with the pulse duration deliberately increased at small wavevectors for increased stability. Cylindrical focusing maximizes the excitation spot size along the wavevector direction, with the spot size selected to generate 100 fringes or more at each wavevector. The large number of fringes in the direction of the wavevector is necessary to avoid walkoff of the propagating acoustic waves. The height of the grating, in the direction perpendicular to the wavevector, is approximately 100 jtm. The experimental wavevector range is 0.08 Lm-' to 1.2 Im-', corresponding to fringe spacings from 74 jm to 5 jLm. All wavevectors are determined by measuring the acoustic frequency in an ethylene glycol sample, for which the speed of sound and its temperature dependence are well known.2 4 At larger wavevectors, the angle is also measured geometrically with a mirror mounted on a rotation stage. The two measurements agree to within 0.001 jtm). A continuous wave argon laser is the probe. Electro-optical gating reduces sample heating. The diffracted signal is detected with a fast photodiode and a digitizing oscilloscope, with 1024 shots averaged for each temperature. For frequencies below 200 MHz, the Antel diode with uniform frequency response to dc is used. At frequencies 300 MHz and higher the Hamamatsu diode acquires the high frequency acoustic component of the data. Data analysis procedures have been described in Chapter 3. 5.3 Results Figure 5-1 shows ISTS data at a small wavevector, in the temperature range where the acoustic, structural, and thermal modes are well separated. Figure 5-2 shows additional data at a higher wavevector. Note the change in the x axis (log scale), and the higher temperature range where structural relaxation can be observed. Over the accessible range of wavevectors in this study, the structural relaxation in glycerol is measured from tens of nanoseconds to hundreds of microseconds. The data are analyzed according to ISTS theory to extract acoustic, structural and thermal diffusion parameters. A Levenberg-Marquardt least squares fit to the time domain signal expression determines the acoustic, thermal, and structural relaxation parameters along with mode amplitudes. ISTS Signal from Glycerol q=0.305 im-' T=350 K cH 1E-3 0.01 0.1 1 10 100 1000 1E-3 0.01 0.1 1 10 100 1000 log Time (ps) Figure 5-1: ISTS data and fits from glycerol at q = 0.305 ptm'. The data show the progression from simple liquid to complex liquid with structural relaxation to simple glass behavior. ISTS Signal from Glycerol q= 1.19 Tm-K T=375 K 1E-3 0.01 0.1 1 10 100 1E-3 0.01 0.1 1 10 100 1E-3 0.01 0.1 1 10 100 H r/ r/c log Time (ps) Figure 5-2: ISTS data and fits from glycerol at q = 1.19 tm 1-. The time axis is considerably shorter than for the smaller wavevector in Figure 5-1. 5.3.1 Acoustics The acoustic frequency and attenuation rate can be determined from a Lorentzian fit to the power spectrum at all temperatures except for the region of maximum damping, where structural relaxation interferes with the spectrum. The acoustic values can also be obtained from a time domain fit to the data. The deviation between the acoustic frequency obtained in the time or frequency domain is less than 0.1%, so there is no advantage to one method over the other. There is a larger difference in the values of the acoustic attenuation, due to imperfections in the frequency response of the detection apparatus or heterodyne effects. Acoustic attenuation cannot be measured accurately at the smallest wavevector due to finite spot size effects. At small wavevectors, the attenuation can also be affected by uncontrolled heterodyning of the signal with scattered light. It is impossible to determine the relaxation time corresponding to the maximum acoustic attenuation with good accuracy. There is considerable uncertainty in the acoustic attenuation measurements near the maximum attenuation. In glycerol, the acoustic oscillations are so strongly damped that only a few oscillations are visible. The data presented in Figures 5-1 and 5-2 show this behavior. In addition, structural relaxation is overlapping in time with the acoustic frequency, and must be included in the fit even though the acoustic and structural relaxation timescales are not well separated. There is also an uncertainty of a few degrees in the temperature where the maximum damping occurs. Speed of Sound in Glycerol 3500 3000- A q=0.305 gm-' A q=0.457 ±m-' * q=0.552 ptm- O q=0.785 gm-' # 2500- q=1.036 gm-' O q=1.186gm-' 2000 - 1500- I 200 220 I 240 260 I 300 I 180 200 220 240 260 280 300 320 I I 360 380 340 360 380 Temperature Figure 5-3: Speed of sound in glycerol at several wavevectors. All wavevectors show a change from high temperature liquid to low temperature glass behavior. The transformation temperature range is a function of wavevector. Acoustic Damping in Glycerol 1200 L 1000 A q=0.305 gm-' A q=0.457 gm" 0 q=0.552 tm-' A 0 q=0.785 gm' A ' O q=1.036 gm-1 q=1.186 Rlm' 800 A 600 cj) 0 400 200 0 180 220 240 340 360 Temperature Figure 5-4: Acoustic damping in glycerol at several wavevectors. The maximum damping occurs at temporal overlap with structural relaxation, and is a function of wavevector. The speed of sound shows the predicted temperature dependence, approaching linear temperature dependence at the high and low temperatures (Figure 5-3). At frequency dependent dispersion occurs, showing the intermediate temperatures, increasing stiffness of the material on the time scale determined by the acoustic frequency. Figure 5-4 shows the acoustic attenuation as a function of temperature. There is a baseline of acoustic attenuation at high and low temperatures due to viscosity and thermal diffusion. The attenuation reaches a maximum at the wavevector-dependent temperature where structural relaxation is occurring on the timescale of the acoustic frequency. There is no evidence of increased acoustic damping at very low temperatures, which would be attributed to the slowing down of the P relaxation into the MHz regime near the glass transition. The acoustic parameters are analyzed as a function of wavevector in Figure 5-5 to confirm the linear dispersion relation for the speed of sound at high and low temperatures. At 340 K in the liquid regime, and at 210 K near the glass transition, the frequency varies linearly with wavevector. At 275 K, where structural relaxation occurs on a time scale close to the acoustic frequencies in ISTS, the frequency does not vary linearly with wavevector. From the linear fit to the data, a slight deviation can be observed. The deviation is very slight due to the limited wavevector range. The acoustic attenuation varies as the wavevector squared at high and low temperatures, where damping is due to thermal dissipation and viscosity (Figure 5-6). The values obtained from the lowest wavevectors are not included in the linear fits at high and low temperatures, as there is substantial error due to spot size or heterodyne effects. At intermediate temperatures where the structural relaxation increases the damping rate, this relation does not hold. Acoustic frequency vs. wavevector 800 700 600 500 400 300 200 - 100 0. 0.0 0.2 0.4 I 0.8 0.2 0.4 0.6 0.8 1.0 1.0 v T=210K * T=275 K . T = 340 K 1.2 1.2 Wavevector (tm 1 ) Figure 5-5: Acoustic frequency vs. wavevector at several At all temperatures, there is a linear temperatures. dispersion relation for the sound wave. 1.4 1.4 Acoustic damping vs. q 300 250 200 * 340 K O 320 K A 275 K v 230 K * 210K A 0 150 A O A 100 A 50 SVV V R 0 n -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 q2 (Pm)-2 Figure 5-6: Acoustic damping vs. q2 at several temperatures. At high and low temperatures there is a linear relationship. The data at intermediate temperatures are affected by structural relaxation. 1.6 1.8 5.3.2 Thermal Diffusivity The thermal decay rate is determined from the long time, single exponential decay of the data. The thermal diffusivity is given by IH/q 2, and should be independent of wavevector. The thermal diffusivity varies weakly with temperature throughout the majority of the supercooled regime (Figure 5-7). As the temperature approaches the glass transition, the thermal diffusivity begins to increase to a value characteristic of the glass state. The temperature where this occurs depends on wavevector, as the material must behave like a glass on the time scale determined by the thermal decay rate. This behavior is similar to the change in behavior of the speed of sound, except it occurs at a temperature much closer to the glass transition as the time scale is much slower. The scatter and the apparent dip in the data at this same temperature range are due to difficulty in fitting the data when the structural relaxation overlaps in time with the thermal diffusion feature. Thermal Diffusivity in Glycerol 0.30 - N q=0.088 mO q=0O.117 gm- 0.25 - A q=0.305 gm-1 > 0.20i a q=0.457 gm- * q=0.552 gm- * " q=1.036 gm <>o q=1.186 gm- 0.15- Ht S 0.10- 0.05 - 180 200 220 240 260 280 300 320 340 360 Temperature Thermal diffusivity vs. temperature in Figure 5-7: glycerol. The increase at low temperatures signal the onset The apparent wavevector of the glass transition. dependence is discussed in the text. 380 There is some deviation from the invariance of the thermal diffusivity with wavevector at the smallest wavevectors. One source of this discrepancy could be inaccuracy in the determination of the wavevector. Another potential source of error at very long times is a sloping baseline due to a intensity variation in the probe pulse. For data where the baseline did not remain flat at zero, the data for the fit were cut off at an earlier time. This could lead to a systematic error in the measurement of the thermal decay rate. Occasionally, at long times and low temperatures, a very long time rise is observed in the signal. In general, this effect decreases in intensity if the probe power is decreased, and is attributed to heating of the sample by the probe. Every effort is made to eliminate this problem, but it is not unusual to still have a non-zero baseline at long times and low wavevectors. Again, the data for the fit are cut off at an earlier time. One of the last two effects is considered the most probable source of the discrepancy in the measurement of the thermal diffusivity at low wavevectors. As discussed in Chapter 3, serious error can result from a contribution to the signal from heterodyning with scattering light. However, it is expected that this would result in low measurement for the thermal diffusion rate, and thus a low value of the thermal diffusivity. As the error shown above is in the opposite direction, it is concluded that this is not the source of the error. In addition, the values obtained for the thermal diffusivity indicate that there is most likely not a large heterodyne contribution to the signal. 5.3.3 Relaxation and Debye-Waller factor measurements The structural relaxation feature appears at intermediate times and temperatures in the data, and is fit to a Kohlrausch-Williams-Watts (KWW) stretched exponential function. The average relaxation times in Figure 5-8 are determined from the time constant FR and the stretching parameter, P. There is a departure from Arrhenius temperature dependence, as shown by the log plot of relaxation times in Figure 5-9. The temperature dependence of the average relaxation time fits a Volger-Tamman-Fulcher (VTF) form (Figure 5-10). The parameters B = 2200 K and TVTF = 133 K agree very well with results from dielectric spectroscopy. 14 Relaxation Times in Glycerol 100 - U U 10 A q=0.305 gm A q=0.457 pm O q=0.119 pm * q=1.036 pm * q=0.086 pm * q=0.552 pm O q=1.186 pm 0.1 220 225 230 235 240 245 250 255 260 265 Temperature Figure 5-8: Relaxation times (log scale) vs. temperature in glycerol at all wavevectors. 270 Arrhenius Fit to Relaxation Times 100 U,- 10 0.1 o.i S------- Arrhenius Fit 0.01 ~1 3.6 3.8 4.0 4.2 4.4 1000/T Figure 5-9: Relaxation times at all wavevectors, with a best fit to Arrhenius temperature dependence. The data deviate from Arrhenius behavior. 4.6 Volger-Tamman-Fulcher Fit to Relaxation Times in Glycerol SIn <,> _- VTF Fit To = 133 K B = 2210 K 15 To=2x10- s A V -2 - -4_ 220 230 240 250 260 270 Temperature Figure 5-10: Vogel-Tamman-Fulcher fit to relaxation times (In scale) in glycerol. Fit parameters are To = 133 K, B=2210 K. Stretching Parameter P in Glycerol 0.70 Sq=0.552 pm q=0.086 gim q=0.119 pm * q=1.036 lm q=0.305 ptm O q=1.186 pm q=0.457 pm ----- p = 0.60 0.65 0.60- 0.55 - 0.50 I 210 ' I 220 ' I 230 * I 240 * I * 250 I 260 I *' 270 Temeprature (K) Figure 5-11: Temperature Dependence of 3 in glycerol at all wavevectors. Error bars are representative for +/- 5 %. The dotted line indicates the average over all wavevectors and temperatures. 280 Within error bars, the values of 3 are independent of temperature throughout the range studied (Figure 5-11). The average for all temperatures and wavevectors is indicated by the dashed line through the data. Dielectric data show a slight increase of 0 from 0.55 at Tg to 0.8 near 300 K. 14 This is not inconsistent with ISTS results, but the experimental error prevents a definitive statement supporting a temperature dependence of p. There is significant scatter in the measurements at different wavevectors. This may be indirectly attributable to the variation of detector response with frequency, as discussed in more detail below in reference to Debye-Waller factor measurements. The quality of the fit to the acoustic damping can affect the fit to the "rise" in the data, which determines the value of p. Figure 5-12 shows only the relaxation function, obtained by subtracting the thermal decay from the square root of the data, as described in Chapter 3. Data at different wavevectors and temperatures can be superimposed by scaling according to the relaxation time, verifying the applicability of time-temperature superposition. Structural Relaxation in Glycerol 0.80.70.60.5 O 0.4•v,= d 0.3 - 0.2I T=230; q=0.119 tm-' 0.1- +tO T=240; q=0.457 m-' 0.0-0.1 I T=260; q=1.036 im-' aI "a II IE-3 II I 0.01 ''''"I1 1 1 ' 0.1 '"aI I 1 I I I III"II I I 10 I I 100 Log Time (ps) Figure 5-12: The relaxation function in glycerol at several temperatures and wavevectors. The data are obtained by subtracting the thermal diffusion mode from the square root of the ISTS signal. The Debye-Waller factor is measured by the relative amplitudes B/(A+B). There is no quantifiable temperature dependence within the error of the data (Figure 5-13) for the entire temperature range of the measurements. There is no quantifiable temperature dependence in any of the individual wavevectors. The line indicating the average magnitude of the Debye-Waller factor at all temperatures and wavevectors is included for reference. In the temperature range studied, we do not observe a square root cusp, or any other anomalous or discontinuous behavior in the temperature dependence. The two smallest wavevectors, where the acoustic frequencies are lowest and the detection response should be the most predictable, are presented separately for clarity at the lowest temperatures (Figure 5-14). The average for these two wavevectors is the same as for all wavevectors within the error bars. Debye Waller Factor in Glycerol 0.75 - 0.70 - 0.65 - qsr+ 0.60 - * q=0.086 m -' * q=0.552 tm- O q=0.119 tm - * q=1.036 A q=0.305 A q=0.457 pm 0.55 220 230 240 250 260 pm - Sq=1.186 m-l --.-DWF=0.66 270I 270 Temperature (K) Figure 5-13: Debye-Waller factor in glycerol for all wavevectors. There is no evidence of a MCT cusp in the temperature range presented. The dashed line indicates the average of all values. 100 m -' 280 Debye Waller Factor in Glycerol 0.700.68 T -r i 0.66 0OO .. .. 0 ...... ----------It.-........... IF --- - .. In 0 0.64- .... ...... --------- O 0.62 0.60 . q=0.086 ptm o q=0.119 tm-' DWF = 0.65 0.58 -.... 215 220 225 230 235 240 245 Temperature Figure 5-14: Debye-Waller factor vs. temperature at the two smallest wavevectors, to highlight the data at the lowest temperatures. The average value is the same as the average for all points. There appears to be some variation in the data with wavevector. Since the wavevectors in ISTS are several orders of magnitude smaller than the wavevectors in neutron scattering where the Debye-Waller factor does vary with wavevector, the possibility that the variation is due to the changing wavevector is rejected. In addition, the acoustic measurements have confirmed that ISTS is in the hydrodynamic regime. Since the acoustic frequency does vary with wavevector, the scatter in the DebyeWaller factor is attributed to a systematic attenuation of the acoustic mode signal by the imperfect frequency response of the detector. In general, if there is attenuation of the acoustic mode signal, the value deduced for A would be artificially small, resulting in an apparent increased value of the Debye-Waller factor. In addition, the frequency response of the detector does not necessarily vary smoothly on the scale of a few MHz, and this can account for some of the error from point to point as well. Use of the two detectors creates additional variation in the data. However, the data at each wavevector can be viewed individually, at least qualitatively. 5.4 Discussion ISTS measurements indicate an effective Debye-Waller factor that is independent of temperature. The magnitude of the Debye-Waller factor is compared to the measurements for salol and CKN (Chapter 4).25 The average value of f0 = 0.66 is significantly larger than the value fo = 0.36 above T, observed for salol. In fact, the amplitude for glycerol is larger than the amplitude for salol even 20 degrees below Tc of the latter. The amplitude is also larger than f = 0.575 in CKN. Recall that the Debye-Waller measures the height of the plateau of the autocorrelation function. The plateau occurs at the crossover between the fast and slow relaxation features. A large Debye-Waller factor indicates that more relaxation occurs through the slow relaxation process. When the Debye-Waller factor is discussed in terms of integrated area of the relaxation spectrum, the Debye-Waller factor is the area of the slow relaxation normalized by the entire integrated area of the both features. The comparison of the Debye-Waller factor amplitudes in glycerol and in the fragile liquids indicates that the amplitude of structural relaxation is larger in glycerol than in these fragile systems. Conversely, it can be stated that fast relaxation is stronger in fragile liquids than in glycerol. It is interesting to observe however, the unusual strength of the boson peak in glycerol, and the difficulty that has presented in the MCT analysis of the susceptibility minimum. A low amplitude of the P relaxation relative to 102 the entire spectrum, which follows from a large Debye-Waller factor measurement, can contribute to the difficulties in the MCT analysis of the high frequency susceptibility spectra. Our results for the Debye-Waller factor indicate that there is no Mode Coupling Theory crossover behavior in the temperature region 225 K to 265 K. This temperature region includes several previously cited Tc values. Tc = 262 K and T, = 248.8 K have been obtained from a power law fits to of the dielectric data. 13,2 1 TC = 225 K has been suggested, although again not convincingly, from light scattering measurements. A range of T c = 223 to 233 K was used in fitting the light scattering spectrum with an expanded MCT coupling term. Several high temperature values, greater than 300 K, also have been reported. To make the most conservative statement, these ISTS measurements only rule out the possibility of a marked anomaly in the temperature region from 228 K to 268 K. It is also unlikely that a cusp similar to those observed in salol and CKN occurs in the temperature range from 270 to 275 K, as we do not observe the square root temperature dependence, or any other significant variation with temperature in this temperature range. The temperature range for these measurements barely extends down to 1.2 Tg = 222 K, where T c is observed for fragile liquids. As MCT predicts a flat Debye-Waller factor above Tc, within the quality of the data, we do not rule out the possibility of Tc at 225 or lower. Efforts are being made to extend the ISTS measurements in glycerol to lower wavevectors and temperatures to address this possibility. If the crossover behavior occurs at a high temperature above 300 K, depending on the amplitude of the temperature dependent term (hq), the data may appear flat within experimental error at temperatures 40 to 70 degrees below the crossover. In fact, the glycerol measurements are not inconsistent with the shape and amplitudes of the fit to salol, with a high Tc. This possibility is also consistent with the large magnitude of the Debye-Waller factor. It is noted, however, that a crossover at high temperatures means that the time scale for relaxation at To is considerably shorter in fragile liquids. From dielectric datal4 at T c = 360 K for CKN, the peak frequency is on the order of 10 kHz, 103 while at 300 K in glycerol the peak frequency is greater than 100 MHz. The frequency ranges covered by the Debye-Waller factor measurements from glycerol and the fragile liquids (Chapter 4) are very similar, so if the process underlying the Debye-Waller factor anomaly occurs as a function of a characteristic relaxation time, it would certainly occur in the frequency range covered by the ISTS glycerol data. From a physical standpoint, any relationship between the T, anomaly in fragile liquids and a high temperature T. in glycerol is considered extremely unlikely. A third possibility is that the Debye-Waller factor anomaly is not dramatic enough to be observed within the error of the data. It has been suggested that with the inclusion of activated processes in extended Mode Coupling Theory, the square-root behavior survives, but the cusp is "softened". It has also been suggested that processes involving the breaking and formation of hydrogen bonds obscure the MCT dynamics in glycerol. 26 A more complicated coupling scheme is necessary to describe the light scattering data. 19 At this point, the possible impact of the coupling scheme used in the light scattering analysis on the Debye-Waller factor prediction is not known. It might be assumed that the coupling of more modes and the inclusion of more temperature dependent parameters in the theory might also have the effect of softening the cusp. The possibility of crossover behavior without an observable anomaly in the Debye-Waller factor can be addressed by considering the temperature dependence of the stretching parameter, which is here considered a measure of the shape of the susceptibility spectrum. The temperature dependence of 0 can be used to test the prediction of time-temperature superposition. Idealized Mode Coupling Theory predicts that the slow relaxation scales with temperature above T, and is arrested below To. Extended MCT restores ergodicity below T, by including "hopping" or activated processes. This might imply, however, that since the mechanism for relaxation is different, there would be a change in the shape of the susceptibility spectrum below T. For both salol and CKN, there is observation of a change in the value of P near or below TC. 25 In both cases, 3 decreases, indicative of enhanced stretching, or the inclusion of additional modes in the relaxation. There is no similar observation in the data for 104 glycerol. The lowest wavevectors may appear to show some slight decrease in 3, but this decrease is well within the error bars. Again, extension of ISTS measurements to lower wavevectors and lower times can address this question further. Addressing the possibility of a crossover in the temperature range 223 to 233 K, we revisit the data obtained for the average relaxation time. Mode Coupling Theory predicts a power law temperature dependence of the relaxation time above Tc. The best fit of relaxation times from ISTS to power law temperature behavior results in Tc = 188 K, with an exponent y = 10 (Figure 5-15). The value of y is much higher than common values for fragile liquids, but does not violate the MCT restrictions for the exponents a and b. The scaling temperature obtained is actually very close to Tg = 185 K. These results are actually quite similar to results from ultrasound and specific heat measurements, which cover a wide frequency range, and are closely related to ISTS. 22 Without a wider dynamic range, a free fit to the MCT prediction for the power law temperature dependence of the relaxation time, allowing data below Tc to be excluded, is not attempted. The relaxation time data above 235 K are fit with a fixed Tc of 225 K for comparison with light scattering analysis (Figure 5-16).2 The resulting exponent is y = 4.3, which is closer to the values used in other scaling analyses (Table 51). As expected, the measured times become faster than the power law prediction approaching Tc. However, these results are far from conclusive given the limited range of relaxation times used in this analysis. As the MCT prediction holds for times above Tc, we are unable conduct a similar analysis to address the question of a high temperature To. 105 Relaxation Times in Glycerol MCT Power Law fit MCT power law t=t0 *(T-TC)Y y=10 T = 188 K -2 - -4- 2I 220 230 240 250 260 Temperature Figure 5-15: Relaxation times in glycerol fit to a power law with no restrictions. The TC value is unusually low and the exponent is unusually high. 106 2 270 Relaxation Times in Glycerol - MCT power law fit with y = 4.3 00 00 S 6 A V 0 2.0 2.5 3.0 3.5 ln(T-225K) Figure 5-16: MCT power law fit with T; restricted to 225 K. The exponent y = 4.1 is somewhat larger than the values obtained in light scattering measurements. 4.0 5.5 Conclusions The dynamics of glycerol have been characterized over the time range from nanoseconds to milliseconds. The structural relaxation dynamics are well fit by a KWW stretched exponential at all temperatures where the structural relaxation timescale is well separated from the acoustic frequency and the thermal decay rate. The relaxation times follow a VTF temperature dependence. The relaxation measurements permit determination of the Debye-Waller factor over a 40 degree temperature range from 1.2 Tg to 1.45 Tg. No evidence of a MCT crossover temperature Tc is observed in that range. The shape of the relaxation spectrum, determined by the stretching exponent 3, does not change with temperature. A lower temperature for T, cannot be ruled out from the available ISTS data, but no anomalies are observed in the specific heat spectroscopy study, which cover the temperature range from the lowest ISTS measurement down to Tg.16 It should not be surprising that Mode Coupling Theory analyses of glycerol have yielded conflicting results. There is no consistency in the values of Tc obtained from the different power laws. There is also no reported evidence of a square root cusp in the Debye-Waller factor measured by neutron scattering. 23 In a review of Mode Coupling Theory, 26 it is clearly stated that Mode Coupling Theory is not expected to apply to intermediate and strong glass forming liquid. "The known fragile systems have a simple microscopic structure. It is perhaps a reasonable guess, that all glass formers exhibit a cross over temperature Tc. However, in the complicated systems the strongly temperature dependent mechanisms of bond breaking and of hydrogen bridge formation presumably mask the more subtle effects responsible for Tc." The conclusion reached from these results is that MCT in its current form is significantly limited in its ability to accurately describe the dynamics of a network forming liquid of intermediate strength. MCT predictions alone have been inadequate to describe the minimum in the susceptibility spectrum without complicated additional analysis. 2,9 Incorporation of additional coupling terms seems to improve the ability of 108 the theory to describe the susceptibility spectrum.19 The strong boson peak can be taken as an indication that high frequency dynamics, likely related to the breakage and formation of bonds, are more prevalent in glycerol than in fragile liquids. ISTS cannot directly elucidate the microscopic mechanism underlying structural relaxation. However, all results point to a relaxation mechanism that is less strongly dependent on temperature than in fragile liquids. It is very likely that this process involves the hydrogen bonding that is prevalent in glycerol, and as a result, MCT, which at present does not include this process, fails to describe the behavior of glycerol. The present results along with earlier reports may motivate attempts to generalize the theory sufficiently to include important microscopic degrees of freedom. References 1 F. J. Bermejo, A. Criado, A. de Andres, E. Enciso, and H. Schober, Phys. Rev. B 53, 5259 (1996). 2 J. Wuttke, J. Hernandez, G. Li, G. Coddens, H. Z. Cummins, F. Fujara, W. Petry, and H. Sillescu, Phys. Rev. Lett. 72, 3052 (1994). 3 J. Wuttke, W. Petry, and S. Pouget, J. Chem. Phys. 105, 5177 (1996). 4 M. Soltwisch and D. Quitmann, Journalde Physique 40, 666 (1979). 5 J. Dawidowski, F. J. Bermejo, R. Fayos, R. Fernandez Perea, S. M. Bennington, and A. Criado, Phys. Rev. E 53, 5079 (1996). 6 D. C. Champeney and R. N. Joarder, Molecular Physics 58, 337 (1986). 7 H. J. M. Hanley, G. C. Straty, C. J. Glinka, and J. B. Hayter, MolecularPhysics 62, 1165 (1987). 8 F. Fujara, W. Petry, R. M. Diehl, W. Schnauss, and H. Sillescu, Europhysics Letters 14, 563 (1991). 9 E. Rossler, A. P. Sokolov, A. Kisliuk, and D. Quitmann, Phys. Rev. B 49, 14967 (1994). 109 10 D. A. Pinnow, S. J. Candau, J. T. LaMacchia, and T. A. Litovitz, Journalof the Acoustical Society ofAmerica 43, 131 (1967). 11 D. W. Davidson and R. H. Cole, J. Chem. Phys. 19, 1484 (1951). 12 P. Lunkenheimer, A. Pimenov, B. Schiener, R. Bohmer, and A. Loidl, Europhysics Letters 33, 611 (1996). 13 P. Lunkenheimer, A. Pimenov, M. Dressel, Y. G. Goncharov, R. Bohmer, and A. Loidl, Phys. Rev. Lett. 77, 318 (1996). 14 A. Pimenov, P. Lunkenheimer, and A. Loidl, Ferroelectrics176, 33 (1996). 15 B. Schiener, R. Bohmer, A. Loidl, and R. V. Chamberlin, Science 274, 752 (1996). 16 N. O. Birge and S. R. Nagel, Phys. Rev. Lett. 54, 2674 (1985). 17 J. P. Kintzinger and M. D. Zeidler, Berichte der Bunsen-Gesellshaft , 98 (1972). 18 T. Christensen and N. B. Olsen, Phys. Rev. B 49, 15396 (1994). 19 T. Franosch, W. Gotze, M. R. Mayr, and A. P. Singh, Phys. Rev. E 55, 3183 (1997). 20 P. Lunkenheimer and A. Loidl, J. Chem. Phys. 104, 4324 (1996). 21 A. Schonhals, F. Kremer, A. Hofmann, E. Fischer, and E. Schlosser, Phys. Rev. Lett. 70, 3459 (1993). 22 Y. H. Jeong, S. R. Nagel, and S. Bhattacharya, Phys. Rev. A 34, 602 (1986). 23 W. Petry and J. Wuttke, Transport Theory in Statistical Physics 24, 1075 (1995). 24 A. Duggal, Ph.D. Thesis, Massachusetts Institute of Technology, 1992. 25 Y. Yang, Ph.D. Thesis, Massachusetts Institute of Technology, 1996. 26 Liquids, Freezing, and the Glass Transition,Vol. 1, edited by J. P. Hansen, D. Levesque, and J. Zinn-Justin (Elsevier Science Pub. Co., Amsterdam, 1991). 110 Chapter 6 ISTS Study of Relaxation in Polypropylene Glycol 6.1 Overview Many polymers easily form supercooled liquids To some extent, the behavior of supercooled polymers is similar to other supercooled liquids such as molecular or ionic systems. For example, the viscosity will change dramatically with temperature through the supercooled regime. In some ways, however, the behavior is different. One common difference between supercooled polymers and supercooled molecular liquids is the "double a peak" feature in the dielectric loss spectrum. As discussed for glycerol, and as predicted by Mode Coupling Theory, the typical relaxation spectrum for a supercooled molecular or ionic liquid consists of a high frequency P relaxation, and a lower frequency, strongly temperature dependent a relaxation. The relaxation spectrum of a polymer, which is most frequently measured by dielectric spectroscopy, shows a similar 0 relaxation feature, but in the a relaxation regime, there are often two peaks, commonly referred to as a and a' peaks. In Polypropylene Glycol (PPG) the a relaxation occurs at the same frequency regardless of molecular weight. The frequency of the a' relaxation depends strongly on molecular weight, and is closer to the a relaxation in a lower molecular weight.1 In PPG, the a relaxation is faster than the c' relaxation. Presumably, the main a relaxation represents the motion underlying the glass transition, as the glass transition temperature changes very little with molecular weight. 2 Although there is a difference in absolute frequency between the features at moderate temperatures, the temperature dependence of the two features seems to be the same, at least for some temperature range. At low temperatures, there is a change in behavior, and the a' relaxation slows down more quickly as the polymer cools. It is unclear whether or not the two features will merge in the high temperature, as high frequency dielectric results are not available. The increased complexity of the relaxation spectrum provides an additional challenge to MCT analysis of polymers as supercooled liquids approaching the glass transition. It must be clearly stated that the oa nomenclature used to describe the slow feature in the dielectric spectra of polymers has its origins in the field of dielectric measurements. It is not completely obvious that there is a direct connection between features called oa and P in dielectric spectroscopy and the slow and fast processes in Mode Coupling Theory. In general, caution should be taken in the comparison of experimental results, as the language used to describe the features in the relaxation spectrum is not entirely consistent in the literature, particularly between the fields of condensed matter physics and polymer science and engineering. MCT does not specifically predict the entire shape of the slow relaxation regime. The KWW function has commonly been found to fit the data, but is not directly derived from MCT. A double ca peak, or a two stage slow relaxation, is not immediately inconsistent with all predictions of the theory. A minimum may still observed between fast and slow relaxations. The lower frequency side of the minimum, corresponding to the high frequency wing of the a relaxation, may still follow the predicted power law dependence. Equivalently, the initial part of the decay of the autocorrelation function from the plateau value may follow the predicted power law behavior. Predictions about the fast relaxation feature may still apply. The Debye-Waller factor prediction may still have the same meaning with a more complicated a relaxation spectrum. Recall that in the idealized theory, the Debye-Waller factor is a measure of how far the relaxation decays from the normalized value 4(O) = 1, to the plateau value which persists indefinitely below Tc. If the Debye-Waller factor prediction is viewed as describing decreasing strength of the fast relaxation processes as the temperature is reduced, it becomes apparent that the particular characteristics of the a relaxation are not pertinent, particularly as the oa relaxation theoretically disappears entirely below Tc in the idealized version of MCT. 112 Although theoretically the double a relaxation may have little relevance to the MCT predictions, there may be experimental implications in testing the predictions. To measure the Debye-Waller factor by integrating the spectrum of the a relaxation, the experiment must have sufficient dynamic range to completely include both a relaxation features throughout the required temperature range. More serious difficulties are encountered in some of the scaling law predictions. A single characteristic frequency or relaxation time of the a relaxation is not now well defined, particularly in the time domain. The two a relaxation features both depend strongly on temperature, but not necessarily in the same way. Clearly, there is now some difficulty in selecting the data for fits to the power law prediction for the temperature dependence of the relaxation time. Possible interpretations are to use either of the two peak frequencies, or an effective peak frequency for the entire a regime. Testing of predictions regarding the minimum of the susceptibility spectrum is unaffected, as the minimum only includes the high frequency wing of the slow relaxation. The principle of time-temperature superposition should be carefully considered. If time-temperature superposition holds, the overall shape of the slow relaxation spectrum scales with temperature and time independently. This implies that all aspects of the double peak spectrum must scale with time and temperature, including both peak frequencies, the shapes of both features, and their relative amplitudes. A Mode Coupling Theory analysis of the dynamics in Polypropylene glycol, MW 4000, has been conducted using a combination of long time and high frequency light scattering techniques. 3 The long time data from Photon Correlation Spectroscopy measures the decay of the correlation function (t). Overall, the data are fit to a KWW stretched exponential. However, there are some deviations both at short and long times. There is additional signal intensity at short times, before the data level off for a period of time prior to the KWW decay. This observation is made at low temperatures near or below the glass transition, and is assigned to P relaxation. The time dependence is fit to a power law decay with exponent a = 0.23. If these data represent the MCT fast relaxation, only the extremely long time tail of the feature is observed on the microsecond 113 time scale. A direct comparison to the depolarized light scattering susceptibility, presented in the same study, show a susceptibility minimum at GHz and higher frequencies at all temperatures, with the fast feature at frequencies above the minimum. Deviations from a single KWW decay are also observed at long times. The original P value obtained from the fits is P = 0.39. When the long time data are fit separately from the main relaxation, the 3 values are P = 0.43 for the main relaxation and P = 0.7 for the long time tail. The long time 3 = 0.7 decay is assigned to the same motion underlying the dielectric a' peak, although due to the scatter in the data it is unclear that the KWW function provides an adequate fit. The time scales of tens of seconds are in good agreement with the dielectric data.' A Mode Coupling Theory analysis of the susceptibility minimum is also attempted from a combination of Brillouin and Raman scattering data at frequencies above 1 GHz. The minimum is only completely characterized at high temperatures, as the slow relaxation moves to frequencies much lower than 1 GHz. The values of the exponent a, obtained from a MCT fit to the minimum, are higher than allowed by the MCT restrictions. The values obtained for Tc from power law fits to the minimum susceptibility and frequency are 265 K and 180 K, using the exponents obtained from the high frequency and low frequency measurements, respectively. Clearly, the exponent needs to be determined less ambiguously before T, can be determined well. Neither of these values agree with the value of Tc = 236 K reported from a combination of PCS and viscosity measurements. 4 Polypropylene Glycol is chosen for ISTS analysis for several reasons. motivation is to characterize the relaxation dynamics of a polymeric system. One MCT predictions are tested using the relaxation data. Recall that MCT is not formulated for a complex liquid, let alone a polymer, so any success of the theory in predicting the behavior of the dynamics might be somewhat surprising. ISTS is used as a quantitative experimental method for measuring the relaxation dynamics of a supercooled polymer, and to test the principle of time-temperature superposition. 114 While time-temperature superposition is incorporated into MCT, it originated empirically as part of the discussion and treatment of relaxation in polymers. 6.2 Experimental Polypropylene glycol (PPG) was purchased from PolySciences (MW 4000) and Aldrich (MW 425). Polypropylene glycol MW4000 (PPG4000) was dried at 100 degrees C under a light vacuum for 24 hours. The warm polymer was forced through a 0.2 tm filter appropriate for high temperature liquids. The liquid was transferred into a tefloncoated aluminum cell with moveable windows, and an RTD immersed in the liquid. To avoid thermal degradation, the lower molecular weight sample was dried at 50 degrees C in a vacuum oven for several days. It was similarly filtered and transferred to a sample cell. The same samples were used to collect all data. The other experimental details are the same as described in Chapter 3 and in the experimental section of Chapter 5. For PPG4000, only the Antel diode was used to collect the data. The Hamamatsu diode was used for the highest wavevectors for PPG425. 6.3 Results Figures 6-1 and 6-2 present ISTS data from PPG425 and PPG4000. At all temperatures, the data show a fast acoustic response and a long time thermal decay. At intermediate times and temperatures, a slow-rising feature appears in the data, which represents structural relaxation. The data presented are collected at a single wavevector, but are representative of data at all wavevectors, with the acoustic and thermal timescales varying as q and q2, respectively. 115 ISTS data from PPG4000 E-3001 0.1 1 10 100 1T=340K00 1E-3 0.01 0.1 1 10 100 1000 1E-3 0.01 0.1 1 10 100 1000 T=230K 1E-3 0.01 0.1 1 10 100 1000 1E-3 0.01 0.1 1 10 100 1000 Log time (pts) Figure 6-1: ISTS data from PPG4000. The data show the growth of the structural relaxation feature as the sample cools. The feature disappears from the data as the relaxation slows beyond the thermal diffusion time. 116 ISTS data from PPG425 T=320 K 1E-3 100 ... 0.01 T=270 K i I. . ... .. I . .. I. I .I . .. I..II . . . . . ... I . . .. I . . I IE-3 0.01 0.1 1 10 100 1E-3 0.01 0.1 1 10 100 Log Time (ps) ISTS data from PPG425, showing the Figure 6-2: structural relaxation feature appearing at intermediate temperatures. 117 . . . . . 6.3.1 Acoustics The measured speed of sound in PPG ( Figures 6-3 and 6-4) compares very well with the velocity of sound obtained from Brillouin Scattering (BS) measurements, taking the difference of wavevector into consideration.5, 6 ISTS and BS data both show a temperature region with dispersion, with a lower temperature range for ISTS due to its lower frequency and wavevector ranges. From the data at high temperatures, ISTS can measure the slope of the zero frequency speed of sound, dVo/dT. The ISTS result averaged for all wavevectors is -3.4 (m/s)/K, and at the lowest wavevector only is 3.3(m/s)/K. This result is higher than the BS measurement of -2.1 (m/s)/K, but due to the high BS frequencies, Vo in the BS experiment consists of only a few points, at higher temperatures than the ISTS values. 6 Also, the cited data were later corrected for a calibration error, which might also have some impact on the measured slope. 5 The analogous infinite frequency, low temperature slope measured with BS is -4.6 (m/s)/K. There are not sufficient data at high enough frequencies from ISTS for comparison. 118 Speed of Sound in PPG425 2800 m q=0.261 Jm 26002400 - 22002000 O q=0.585 tm' A q=0.817 Pm- A q=1.36 m-' - 18001600 1400 1200180 200 220 240 260 280 300 320 340 Temperature (K) Figure 6-3: wavevectors. Speed of sound in PPG425 at several 119 360 Speed of Sound in PPG4000 28002600 2400 2200 2000 1800 - * q=0.169 tm O q=0.485 tm * q=1.21 CVm- 1600 1400 - 12001000 lI ' 180 I 200 220 240 260 280 300 320 340 360 Temperature (K) Figure 6-4: Speed of wavevectors. sound in PPG4000 at several 120 380 Comparison of Speed of Sound in PPG425 and PPG4000 2800 2600- Oo0 0 2400- 22002000 - 1800- * PPG425 q=1.36 tm-' O PPG4000 q=1.21 tm - 1600140012001000 I ' ' I ' I ' I ' I ' I ' I ' I ' I ' I . 180 200 220 240 260 280 300 320 340 360 380 Temperature Figure 6-5: Comparison of the speed of sound in PPG425 and PPG4000. The wavevectors are similar, but not identical. The speeds of sound in PPG4000 show virtually identical behavior to PPG425. Figure 6-5 presents data from both molecular weights as a function of temperature. The data cannot be compared quantitatively, as the wavevectors are different for the two samples, and there may be small difference in the material density. Brillouin Scattering measurements also report that the speed of sound at fixed q does not vary with the molecular weight in PPG, considering polymers of molecular weight 425, 1025, 2025, and 4000. 5 It is interesting to note that the speed of sound is similar despite a very large different in the viscosities of the different molecular weight polymers at room temperature. Acoustic Damping in PPG425 700 600 - r:ri r mm 500 - o # 400 - o q=0.585 gm1- 4-. Ct 0E 300 S , Ct q=0.261 tm - * O U* 0 0 o o A0 200 AAA A q=0.817 pm A q=1.36 tm-i AOM 100A 4 0 OAA 0 ' I 220 ' 240O I I I I I 260 200 ' 260 2 lO ' IO 280T ' as 4 I 300 ' I ' 320 Temperature Figure 6-6: Acoustic damping in PPG425. There is a main peak at all wavevectors, and some indication of a shoulder on the low temperature side of the maximum. I 340 - Acoustic Damping in PPG425 150 aa 120 c A q=1.36 tm - rCl . , C 30 - A 220 A A aA 240 260 280 300 320 340 Temperature Figure 6-7: Acoustic damping in PPG425 at one wavevector, highlighting the low temperature shoulder. The acoustic damping rate in PPG425 passes through a maximum at the temperature where the relaxation timescale coincides with the acoustic frequency. In PPG425, at temperatures lower than those corresponding to the maximum acoustic damping, there is some indication of a second peak, or a shoulder in the damping rate. Figure 6-7 shows the wavevector with the clearest indication of this behavior. A similar feature is observed in measurements of the derivative of the real part of the longitudinal modulus at constant wavevector measured by Brillouin scattering. 7 The Brillouin data are analyzed in terms of Debye relaxation, and no insight is offered on the physical origin of the low temperature deviation from the Debye fit. Other Brillouin scattering data do not show a low temperature feature. 5 Acoustic Damping in PPG4000 200AA rj 150- S A AAA A A A a 100- a S S q=0.967 im AA 50- AZA aAa ap zzzzz~z C3 nAaA 180 200 20 240 260 28' I 0 0 3 40 360 380I I 180 200 220 240 260 280 300 320 340 360 380 Temperature Figure 6-8: wavevector. Acoustic damping in PPG4000 at one - The acoustic damping in PPG4000 is presented in Figure 6-8. Similar to PPG425, there is a possibility of a low temperature feature in the damping, but in this case there are not sufficient data to draw any conclusions. The quality of the data at other wavevectors is poorer. Previous results from PPG4000 also give an extremely subtle indication of a second feature in the acoustic damping.8 As with PPG425, this behavior is not observed in Brillouin Scattering measurements of acoustic attenuation. 5 6.3.2 Relaxation Measurements The average relaxation time is obtained from the fit to the structural relaxation feature in the ISTS data. Over all wavevectors, relaxation times can be measured from tens of nanoseconds to tens of microseconds. For both PPG molecular weights, this corresponds to observation of structural relaxation for temperatures 25 to 50 degrees above the glass transition. Figure 6-9 presents the average relaxation times in PPG425 on a log plot, with PCS results included for comparison. The times from ISTS do not appear to extrapolate at low temperature to coincide with measurements from PCS. As it is unclear whether the two techniques probe the same dynamics, no conclusion can be drawn from the discrepancy. Also, the relaxation time is changing very rapidly with temperature in this regime, and error in temperature measurements may cause deviations in the relaxation time comparison. There also may be differences in the sample, such as water content. 125 Average Relaxation Time in PPG425 10000 - q=0.261 tm-1 1000 q=0.585 tm -' q=0.817 gm- 100 q=1.36 tm-1 ISTS acoustics 10- PCS (VV) Vt- V PCS (VH) 10.1 0.01 210 215 220 225 230 235 240 245 250 255 260 265 270 275 Temperature Figure 6-9: Log plot of relaxation times in PPG425 obtained by ISTS and other experimental methods. The high temperature ISTS data are determined from the maximum in the acoustic damping. ISTS data appear to extrapolate to faster relaxation times at low temperatures, compared to the PCS results. 126 Comparison of Relaxation Times in PPG425 and PPG4000 4 2 PPG425 o 0 PPG4000 V -2 rnrn -4 4.0 4.1 4.2 4.3 4.4 4.5 1000/T (1/K) Figure 6-10: Comparison of Relaxation Times for PPG425 and PPG4000. The data are presented on logarithmic axes. The left and right axes are offset by one order of magnitude for clarity. The scale is the same for both data sets. The dashed lines are Arrhenius fits to the relaxation time. The relaxation time data for PPG425 and PPG4000 are simultaneously presented for comparison in Figure 6-10. The temperature dependence of the relaxation times is non-Arrhenius in this temperature range, as shown in Figure 6-10. The deviation from Arrhenius behavior is more apparent, at the lower molecular weight. This illustrates the need for a wide dynamic range in experiments that address the temperature dependence of the relaxation time. Particularly for a strong or intermediate liquid, relaxation times for only a limited range can appear to have an Arrhenius temperature dependence. 127 The data fit a Vogel-Tamman-Fulcher form for the temperature dependence. The results for PPG425 and PPG4000 are displayed in Figure 6-11 and Figure 6-12. The VTF results for B and To are very similar to results from VTF fits to viscosity data. 9 For the temperature range from 243 K to 233 K, the reported parameters are B = 955 K, To = 174 K for PPG425 and B = 898 K, To = 176 K for PPG4000. VTF Fit to Relaxation Times in PPG425 32- --VTF fit .. 0A V VTF FIT -1 IL M' -2 -'g -3 To= 170 ±5 K B= 1020 +20 K T 5e-14 s -- -4 220 225 230 235 240 245 Temperature Figure 6-11: Vogel-Tamman-Fulcher fit to relaxation times in PPG425. The results are To = 170 K and B = 1020 K. 128 250 VTF Fit to Relaxation Times in PPG4000 2 1 0- I. -1 "'Ii. -2 -3 -4 -1 -4 -3 -2 220 " 225 230 ' I ' 235 230 " 240 i2 245 250 Temperature Figure 6-12: VTF fit to relaxation times in PPG4000. Fit results are included on the Figure. The relaxation times are also fit to a power law temperature dependence for comparison with Mode Coupling Theory predictions (Figure 6-13). When relaxation times from all temperatures are included in the fit, the power law fit results in Tc = 205 K. When only higher temperature data are included in Figure 6-14, above 235 K, the Tc measured is 217 K. As expected, the fit restricted to higher temperatures seems to 129 provide a better representation of the highest temperature data. It is expected that further restrictions in the temperature range of the data would result in progressively higher values of To. Since there is no obvious point at which to begin restricting the data included in the fit, it is impossible to extract a meaningful measure of Tc from the relaxation time data. It is noted that the higher temperature T, is accompanied by a value of the exponent that is more typical of MCT fits. MCT Power Law Fit to PPG425 43- 2 1 m • <R> -- MCT Power Law Fit o "mm s S MCT Power Law fit - y= 7.3 = 205 K -1 VT - -2 - -3 -4 220 225 230 235 240 245 Temperature Figure 6-13: Average relaxation time in PPG425 fit to a MCT power law for the temperature dependence. All data are included in the fit. 130 250 MCT Power Law fit to PPG425 Selected Temperature Range 4- *~ 3U 2- *~~ m , 1- <T R> i =L V -s ----- 0- MCT Power Law Fit •MCT Power Law Fit y = 4.3 -1 Mmk v T= 217K T > 230 K 'I -2 Ui a m'm. . -3 -4 I~ I 220 ~ * 225 230 I 235 240 245 250 Temperature Figure 6-14: Average relaxation times, with data at temperatures above 230 K fit to a MCT power law temperature dependence. The intermediate time, intermediate temperature data representing structural relaxation are fit to a Kohlrausch-Williams-Watts stretched exponential function. The value of 0 reflects the degree of stretching, or the extent of the deviation from Debye behavior, where P is 1. The measurements of P for PPG2425 show some systematic variation with temperature, with p decreasing with decreasing temperature (Figure 6-15). There is some scatter in the data at different wavevectors, but each wavevector shows a similar temperature dependence. The values appear to decrease to coincide with the P values obtained from PCS in the VH geometry (Figure 6-16). 9 It is also noted that the magnitude of 3, near 0.5 for the temperature range from 230 to 240 K, is similar for the two molecular weights. This is taken as further evidence that the structural relaxation observed in ISTS is similar for both molecular weights. Stretching Parameter 1 in PPG425 0.75 0.70 T - 0.65 A A" - aa 0.60- A 0.55 *,E 0.50 0.45 0 0.40- 0.35 * q=0.261 ttm- o q=0.365 rtm' 9 q=0.496 im' O q=0.585 tm- A q=.817 [m' A q=1.36 pm -' I 220 225 235 230 240 245 Temperature Figure 6-15: Stretching Parameter P from PPG425 as a function of temperature at several wavevectors. 132 250 Stretching Parameter 3 in PPG425 0.75 - 0.70 0.65 - A AX 0.600.55 " q=0.261 gm ' 0.50 - q=0.365 gm' q=0.496 .m' 0.45 q=0.585 pm - " q=0.817 gm') q=1.36 gm" 0.40- PCS (VV) + PCS (VH) 0.35 210 210 215 I 215 20 220 ' 230 25 225 Temperature 240I 235 240 I 245 Figure 6-16: Stretching Parameter P from PPG425 as a function of temperature at several wavevectors, including measurements from VV and VH Photon Correlation Spectroscopy. 133 250 Stretching Parameter 3 in PPG4000 0.55 - 0.50- a * 4-j A! M d) S 0.45 - oJ2 Q: 0.40- 0.35 2I 220 * q=0.169 tm A q=0.485 jm -1 * q=1.21 tm-' ' I I ' 225 230 I 235 ' 2 240 2 245 Temperature Figure 6-17: Stretching parameter 3 in PPG4000 at several wavevectors. There is a decrease in the value of 3 with decreasing temperature. The Debye-Waller factor in PPG425 shows some slight variation with temperature, with higher values at lower temperatures. In Figure 6-18, the temperature dependence is obscured by the scatter among different wavevectors. In Figure 6-19, the data obtained at the two highest wavevectors, where data from two detectors are combined, are rescaled for better agreement with the measurements at lower wavevectors in the temperature range where data from multiple wavevectors could be analyzed. It is not improbable that in the merging of data from two detection systems, a systematic error is incurred in the relative amplitudes of the acoustic and structural relaxation modes. However, it is stressed that there is no rigorous basis for this adjustment. The Debye-Waller factor in PPG4000 is presented in Figure 6-20. As only one detector was used to obtain the ISTS data for PPG4000, no rescaling of the Debye-Waller factor is attempted. The overall magnitude of the temperature variation of the DebyeWaller factor is similar for the two molecular weights. Debye Waller Factor in PPG425 0.80 0.75 - q=0.261 pm q=0.365 lpm 0.70 - q=0.496 lpm" q=0.585 ipm- 0.65 - 1 q=0.817 ipm q=1.36 vpm 0.60 0.55 - U AI 0.50 0.45 0.40 - 220 225 225 235 230 240 245 250 Temperature Figure 6-18: The Debye-Waller Factor in PPG425 as a function of temperature. There is no distinct evidence of the anomaly predicted by Mode Coupling Theory. 136 Debye Waller Factor in PPG425 0.70 (rescaled high wavevectors) q=0.261 Ltmz 0.65 - q=0.365 lam ' q=0.496 lm' q=0.585 lam' 0.60 E 0 q=0.817 m-'m 0o q=1.36 lpm-' *) -1 L -..- 0.55 A 0.50 0.45 220 225 230 235 240 245 Temperature Figure 6-19: The Debye-Waller Factor in PPG425 at different wavevectors. The measurements from the highest wavevectors are arbitrarily rescaled to 90% to better agree with the measurements at lower wavevectors. 137 250 Debye Waller Factor in PPG4000 0.70 - 0.65 - * q=0.169 tm' O q=0.485 pm - A q=1.21 pm' 0 0.60- 0 iga _L 0.55 0.50 0 - 12 225 - 230 235 240 Temperature Figure 6-20: Debye-Waller factor measured for PPG4000. 245 6.4 Discussion In ISTS, there is no compelling evidence of the second, a' feature as a structural relaxation mode. Although the available dielectric data cover a wide frequency range from a few hertz into the megahertz range, higher frequency data are not available. Brillouin scattering is conducted at higher frequencies, but the range is somewhat limited and does not map out all of the features in the spectrum. In particular, the behavior of the a and a' features at high temperatures is not available. We can only assume that the features converge at high temperatures, and separate at some point, or that one feature is not observable in light scattering or ISTS. As the liquid cools, the a' relaxation slows down more rapidly than the a relaxation. The a' relaxation depends on molecular weight. It follows from this scenario that in a higher molecular weight polymer, where the a and a' peaks are more separated at lower temperatures, the a' relaxation splits off from the main a peak at higher temperatures. Note that this scenario implies deviation from time temperature superposition at high frequencies. In ISTS measurements, the timescale where structural relaxation can be observed is that where the relaxation time is slower than the acoustic frequency but faster than the thermal decay. Acoustic damping provides evidence of structural relaxation at times moderately faster than the acoustic frequency. The acoustic frequency increases by about a factor of 2 between the liquid and the glass, and the thermal diffusion rate varies only weakly with temperature until very close to the glass transition. Thus, the time where structural relaxation is visible is nearly the same for all temperatures at a give wavevector. Any structural relaxation feature that contributes to the density dynamics in the experimentally defined time window can appear at any temperature. To completely analyze ISTS results for the presence of two relaxation features, a semi-quantitative comparison is made to dielectric results. Exact agreement between dielectric peak frequencies and ISTS relaxation times is not expected, but the order of magnitude can be used for comparison to the ISTS results. Dielectric peak frequencies are reported for PPG4000 and PPG880.1 It is assumed that frequencies for PPG425 will 139 follow the molecular weight trend, and be the same for the ca relaxation and slightly faster for the a' relaxation. At T = 235, the a peak frequency is 100 kHz for both weights, and the a' peak occurs at 10 kHz for PPG880 and 100 Hz for PPG4000. At T = 222 K, the features have slowed down to 3 kHz for the acpeak and 200 Hz and 3 Hz for the PPG880 and PPG4000 a' peaks. The relaxation times from ISTS are in agreement with the a relaxation frequency within an order of magnitude. For ISTS, there are temperatures and wavevectors where both relaxations in PPG425 should be visible if both motions couple to the density dynamics. Presumably, the effect of two features on the ISTS data would be a poor KWW fit to the structural relaxation feature. However, it is noted that the sum of two Kohlrausch functions can still be well represented by a Kohlrausch function if the characteristic relaxation times are closely spaced. The widest separation of the features is at lower temperatures. At the lowest wavevector, where the time window for structural relaxation is widest, the data have been closely examined for deviations from the KWW fit. No extraordinary discrepancies are observed, including when the thermal decay is subtracted from the data and the relaxation function is observed directly. At temperatures approaching the glass transition, there is occasionally a very long time rise to the data mixed with the end of the thermal decay. On an experimental basis however, this can be attributed to a variation of the probe intensity with time. If the a' feature were observed in the ISTS experiment, it could conceivably be observed at higher temperatures, particularly in the PPG4000 data. The a and a' frequencies are well separated at the higher molecular weight at low temperatures. Presumably, at some temperature the two features converge, implying that at an intermediate temperature, the separation is one or two orders of magnitude. If this condition occurs when the main a relaxation is faster than the acoustic frequency, only the a' relaxation would be observed as structural relaxation. This behavior is not observed at any temperature or wavevector. There is some evidence of a second relaxation feature in the acoustic damping rate as a function of temperature. The initial interpretation of this result in ISTS is that a 140 shoulder in the damping rate is indicative of a second maximum in the relaxation spectrum. This second maximum would actually be a higher frequency feature in the relaxation spectrum measured at constant temperature. When considering the acoustic damping rate measured at constant wavevector, "lower" frequencies features appear at higher temperatures, and a "higher" frequency feature appears at lower temperatures, where it has slowed down to the acoustic frequency timescale. The ISTS damping rate data are not of sufficient quality or quantity to make numerical measurements of relaxation times from acoustic damping. However, estimates are that the main damping feature occurs near 400 MHz at a temperature near 270 K. The second feature also has a frequency near 400 MHz at a temperature near 255 K. Numerical values for dielectric peak frequencies are not available in this regime for comparison, however, both the a and a' frequencies are below 1 MHz at 255 K.1 From extrapolation of the dielectric results, the main peak in the acoustic damping rate is assigned to the a relaxation. A peak due to the a' relaxation would occur at higher temperatures, and is not observed. Further information cannot be gained from the Brillouin Scattering measurements, as the frequency range is over an order of magnitude higher, and the acoustic damping displays a broad maximum at temperatures greater than 300 K. If there is a second feature contributing to the damping at higher frequencies and temperatures, it is unlikely that the two features could be distinguished, as they likely merge at high temperatures. The breadth of the acoustic absorption spectrum, however, does not contradict the possibility of multiple, or complex relaxation processes. The recurrence in the damping occurs at a temperature below the peak, so it is due to a process that occurs at higher frequencies. Thus, this behavior is not assigned to the a' relaxation as observed in dielectric spectroscopy. If anything, the additional damping should be assigned to the slowing down of a "p-like" relaxation into the ISTS frequency regime. Also, the frequency for this feature does not change nearly as rapidly with temperature as the temperature dependence of the a relaxation. The frequency remains in the hundreds of MHz for about 30 degrees in PPG425. There is no evidence of a feature in the hundreds of KHz to 1 MHz regime in dielectric data at temperatures down to 209 K. l 0 No similar shoulder is evident in the acoustic damping data for glycerol (Chapter 5), salol, or CKN."1 Although this seems somewhat slow for P relaxation as reported for non-polymeric materials, it is consistent with PCS results reporting the tail of the 3 relaxation on a microsecond time scale. 4 This would also imply that the 3 relaxation in PPG has a different character than in the non-polymer glassforming liquids. The description of this high frequency feature should not be considered synonymous with the MCT fast relaxation. In fact, the susceptibility minimum, which is analyzed in terms of MCT predictions, occurs at frequencies greater than 1 GHz for all temperatures above the glass transition. 3 The most definitive statement regarding the a and a' relaxation that can be made from ISTS data is that if the a' relaxation is strongly coupled to density fluctuations, it is indistinguishable from the a relaxation on the timescales from nanoseconds to microseconds in both molecular weights. There is also the strong possibility that although the motion associated with the a' feature is strongly associated with a dipole moment in the polymer, it does not make a strong contribution to the density dynamics. If the a' relaxation is not coupled to the density dynamics, the MCT analysis of the relaxation dynamics in a polymer is simplified. To date, we are not aware of the observation of two slow features in an experiment that directly probes the density dynamics. There is some evidence in the PCS measurements of a very slow relaxation near the glass transition,3 but the relationship between PCS data and density dynamics is not completely clear. With one slow contribution to the density dynamics, the relaxation in the polymer follows the MCT scenario of a fast and a slow process. The fast process is represented by the motions contributing to the ISTS acoustic response, and is above the frequency range of the dielectric experiments on the polymer in the liquid state. The Debye-Waller factor will still be measured by the relative amplitudes of slow, structural relaxation. Time-temperature superposition will still be indicated by a temperature- invariant value of the stretching parameter 0, which represents the width of the relaxation spectrum. 142 The relaxation time measurements in both molecular weights of polypropylene glycol do not adhere well to the MCT predictions. Due to limited dynamic range, the power law fits of the relaxation time to the power law prediction are inconclusive. A power law adequately describes the temperature dependence of the relaxation time, but a value of TC cannot be reliably extracted from the data. A few comments need to be made regarding the viscosities and the relative fragilities of the different molecular weights. At room temperature, PPG4000 is several orders of magnitude more viscous than PPG425. Traditionally, the glass transition temperature occurs when the viscosity reaches a value of 1013 poise. Considering these two endpoints, PPG425 would be more fragile than PPG4000, as the viscosity must undergo a more dramatic change between room temperature and the glass transition. The T, values reported, however, rely on calorimetric measurements 2 , and not on a measured viscosity. In the range accessible through ISTS, the relaxation times show a very similar temperature dependence. The VTF parameters are also very similar. The Arrhenius fit to the relaxation times for PPG4000 is slightly better. However, this kind of judgment is difficult to make without an extremely wide dynamic range. To our knowledge, there are no measurements of the viscosity, or the relaxation time over the entire range from room temperature to the glass transition. From the evidence available, we assume that PPG425 is more fragile than PPG4000, but the difference is not dramatic. The difference in the fragility is not as large as between a fragile liquid such as salol, and an intermediate liquid such as glycerol. The fragility of both molecular weights is somewhere between these materials. There is no observation of an anomaly in the Debye-Waller factor in the temperature region accessible through ISTS, however, in both molecular weights, there is some temperature dependence. The slight temperature dependence could be an indication of several possibilities. The decreasing Debye-Waller factor as the material warms could simply be an extension of the glassy behavior into the supercooled regime, as observed in neutron scattering. 12 The ISTS measurements cannot rule out the possibility of a crossover temperature above the temperature range where the Debye-Waller factor measurements are made. A steadily decreasing value of P is observed in both molecular weights, which would be somewhat similar to observations in fragile glass forming liquids. The temperature range in this study does include 1.2Tg, 240 K for both polymers, and there is no clear anomaly at that temperature. Power law MCT fits have resulted in TC = 236, 4 which also is not supported by ISTS measurements. Of course, the possibility that the anomaly is softened to the point of not being observable always exists. There is no evidence of a distinct anomaly of any sort in the Debye-Waller factor in either molecular weight. The possibility remains that there is a MCT square-root cusp in the Debye-Waller factor at temperatures above the temperature range presented. The scatter in the data prevents quantitative analysis of the temperature dependence, such as a square root dependence. The data for PPG425 are only suggestive of this behavior with the rescaled wavevectors. The strongest conclusion that can be reached is that there is evidence of some increase of the Debye-Waller factor as the material cools. The absolute magnitudes of the Debye-Waller factor are similar for the two molecular weights. The evidence from the relaxation times and stretching parameter 0 already indicate that a similar motion underlies the structural relaxation observed in ISTS for both weights. Similar Debye-Waller factors indicate that the relaxation has similar strength in both molecular weights. The temperature dependence in the Debye-Waller factor mirrors the temperature dependence of P. A smaller value of P indicates more stretching in the relaxation spectrum. This may naturally imply a larger integrated area of the relaxation spectrum, i.e. a larger Debye-Waller factor. Thus, an increase in the Debye-Waller factor might be consistent with a decrease in 3, provided the integration limits completely include the relaxation feature. The values of P for PPG4000, particularly at q = 0.169 tpm' show a clearer decrease with T. In the case of PPG4000, this is most likely due to an increased width in the a relaxation at low temperatures. The a' relaxation at these temperatures is slower than the thermal diffusion, even at this small wavevector. Only the tail of the a' relaxation would fall in the ISTS time window, and would only have an impact on the 144 shape of the structural relaxation if the motions strongly contribute to the density variation. These measurements do not agree with PCS results, which report a KWW behavior with P = 0.20 at 242 K. At decreasing temperatures, there appear to be two features. By T = 208 K, the data revert to a single KWW decay with P = 0.39. This value is in better agreement with the ISTS results. As the experiment only shows the long time tail of the decay at high temperatures, it is possible that the low value is reflective of two features that are closely spaced or overlapping in time. There is a striking similarity between both the Debye-Waller factor measurements and the 0 values obtained for the two molecular weights. The relaxation times also nearly coincide, given the slight difference in the glass transition temperature. For these results, we can conclude that the origin of the structural relaxation mode in ISTS is closely related to the main a peak in dielectric loss, which has also been found to be independent of molecular weight. By comparison of the dielectric frequency with the ISTS relaxation time, it is also clear that the main a relaxation is the mode that is active in ISTS. For several of the measurements made in PPG, there appears to be some systematic variation of the measurement with wavevector, despite expectations that the value should be independent of wavevector. In the zero wavevector regime of ISTS, the Debye-Waller factor, the relaxation time, and the stretching parameter P should be independent of wavevector. There are several sources of error that can be introduced in an explanation of the apparent wavevector dependence. The frequency response of the detector is one candidate to explain variation of the Debye-Waller factor, relaxation time, or P with wavevector. As the wavevector changes, the signal moves to a different area of the frequency response curve of the detection system. Also, at the highest wavevectors in PPG425, two different detectors may be used to completely span the frequencies in the signal. Despite every effort to accurately account for bandwidth characteristics, error introduced in these processes cannot be totally eliminated. In the case of a polymer that is known to exhibit a complex and stretched relaxation spectrum, the variation with wavevector may be a result of the experimental 145 limitations of ISTS. As the wavevector is changed in the ISTS experiment, the frequency "window" for observation of the structural relaxation changes as well. At the high frequency end, the acoustic frequency increases linearly with wavevector. On the low frequency end, the thermal diffusion rate varies as wavevector squared. The combination of these two conditions leads to a wider time or frequency window at low wavevectors than at high wavevectors. When the Debye-Waller factor is considered as an integration measurement, a larger frequency window for the integration would result in a larger integrated area, and thus a higher Debye-Waller factor. 6.5 Conclusions ISTS measurements have characterized the structural relaxation dynamics of PPG425 and PPG4000. The relaxation times, relaxation shape, and Debye-Waller factors are quantitatively similar for both molecular weights. Thus, we conclude that the structural relaxation mode that is active in ISTS, i.e. coupled to the density dynamics, is most closely related to the main a feature in dielectric spectroscopy. MCT predictions do not adequately describe the relaxation dynamics. There is no evidence of a crossover temperature in the range studied, either by an anomaly in the Debye-Waller factor measurement or by a divergence of relaxation times. There is some enhanced stretching at low temperatures, but it is impossible to determine if this is a result of some contribution from an additional feature, or an indication of crossover behavior. As with glycerol, we conclude that Mode Coupling Theory in its current form is not adequate to describe the dynamics of a polymer liquid. 146 References 1 A. Schonhals and E. Schlosser, Physica Scripta T49, 236 (1993). 2 G. P. Johari, A. Hallbrucker, and E. Mayer, Journalof Polymer Science, Polymer Physics Edition 26, 1923 (1988). 3 R. Bergman, L. Borjesson, L. M. Torell, and A. Fontana, Phys. Rev. B 56, 11619 (1997). 4 D. L. Sidebottom, R. Bergman, L. Borjesson, and L. M. Torell, Phys. Rev. Lett. 68, 3587 (1992). 5 C. H. Wang and Y. Y. Huang, J. Chem. Phys. 64, 4847 (1976). 6 Y. Y. Huang and C. H. Wang, J. Chem. Phys. 62, 120 (1975). 7 H. H. Krbecek, W. Kupisch, and M. Pietralla, Polymer 37, 3483 (1996). 8 A. R. Duggal and K. A. Nelson, J. Chem. Phys. 94, 7677 (1991). 9 C. H. Wang, G. Fytas, D. Lilge, and T. Dorfmuller, Macromolecules 14, 1363 (1981). 10 K. Pathmanathan, G. P. Johari, and R. K. Chan, Polymer 27, 1907 (1986). 11 Y. Yang, Ph.D. Thesis, Massachusetts Institute of Technology, 1996. 12 W. Petry and J. Wuttke, Transport Theory in StatisticalPhysics 24, 1075 (1995). 147 Chapter 7 Summary and Future Directions ISTS studies have been conducted on glycerol and Polypropylene Glycol of two molecular weights. The relaxation dynamics have been characterized as a function of temperature for times from a few nanoseconds to hundreds of microseconds. On a qualitative level, these liquids behave similarly to fragile liquids. Three hydrodynamic modes are observed, a fast acoustic mode, slow thermal diffusion, and temperature dependent structural relaxation. In glycerol, the results are analyzed according to Mode Coupling Theory predictions. The Debye-Waller factor is measured for a temperature range from 1.2 Tg to 1.45 Tg, and no evidence of a MCT square root cusp is observed in this range. The stretching parameter, P, is constant with temperature in this same range, indicating that time-temperature superposition holds. temperature dependence. The relaxation times are fit to a power law A fit to all the data results in a To value below the glass transition temperature. When the data are restricted to high temperature values, the data are not inconsistent with TC = 225 K, but this should not be considered a definitive measurement of T. Without data at longer times and lower temperatures, we are unable to rule out the possibility of a crossover temperature below 225 K. However, due to the conflicting evidence from ISTS and other experiments, we conclude that Mode Coupling Theory is very limited at present in its ability to explain the dynamics of a material with complex interactions, such as the hydrogen bonding that is prevalent in glycerol. ISTS is also used to characterize the relaxation dynamics in two molecular weights of Polypropylene glycol. The dielectric spectrum shows two features in the a relaxation regime. ISTS results only indicate one relaxation process, which is quantitatively similar for both molecular weights. The measured relaxation times, stretching parameters, and Debye-Waller factors are similar for both molecular weights. 148 We conclude that the dynamics observed in ISTS are related to the main a relaxation in dielectric spectroscopy. There are weak observations of a more complicated relaxation scenario, such as decreased values of P at low temperatures, but no quantitative conclusions are drawn. The second relaxation does not appear to be strongly coupled to the density dynamics probed by ISTS. As with glycerol, there is no evidence of a Mode Coupling Theory crossover temperature in either molecular weight. Although the Debye-Waller factor is weakly temperature dependent, there is no clear evidence of an anomaly in the temperature range studied. There is a decrease of P with temperature, but this observation cannot be definitively interpreted as an indication of crossover behavior. Power law fits to the relaxation times cannot clearly determine a To value in the supercooled regime. Mode Coupling Theory applied to Polypropylene Glycol appears to suffer from the same limitations as in the application to glycerol. The dynamics in a material with very complex interactions are not adequately described by the theory. Additional theoretical considerations may help to extend MCT capabilities into this regime. Continuing efforts to apply Mode Coupling Theory to the dynamics of glass forming liquids would benefit from expanding the frequency range of ISTS. Expanding to include higher wavevector measurements can permit analysis of the minimum in the acoustic modulus spectrum. The acoustic modulus can be computed from the acoustic parameters obtained from ISTS and ISBS measurements. This has been done to some extent for salol and PPG, using a picosecond pulsed probe version of the ISTS experiment. 1,2 The difficulty in manually generating large angles can now be bypassed by using phase masks, and the data could be improved with measurements at more wavevectors. Instead of taking data at a range of temperatures at a single wavevector, the sample can be kept at a fixed temperature while the wavevector is changed by switching patterns on the phase mask. This greatly reduces the uncertainty due to temperature variation in the measurement of the modulus. In order to explore the minimum in the susceptibility spectrum, the frequency range of the acoustic measurements made with ISTS or 149 Impulsive Stimulate Brillouin Scattering (ISBS) must be extended well into the GHz regime, to a frequency of 30 GHz or higher. Generating an acoustic frequency of 30 GHz, assuming a speed of sound near 3000 m/s, would require a wavelength of about 100 nm. In order to generate a grating with the appropriate fringe spacing, ultraviolet excitation wavelengths must be used. At present, the technology is not available to generate the necessary excitation pulses. The practical limit for time resolution with current femtosecond technology is sub-100 femtosecond. Frequencies into the THz regime can be observed, if the acoustic wave can be generated. The highest practically generated wavevector would involve the third harmonic of Ti:Sapphire pulses, at 266 nm. This excitation could be used to generate fringe spacings on the order of hundreds of nanometers, corresponding to acoustic frequencies near 10 GHz. Femtosecond excitation and probing would provide sufficient time resolution for these acoustic frequencies. The sample should be transparent to avoid photochemical effects. Assuming sufficient energy in the excitation and probe, ISBS signal could be obtained from transparent samples. A simpler, but limited way of extending the frequency range of ISTS measurements would continue with a cw probe and use a streak camera for detection. Steak camera detection, with femtosecond excitation pulses, can provide time resolution into the tens of picoseconds, extending the frequency range to a few GHz. This improvement would be sufficient for some rudimentary evaluation of the susceptibility minimum at certain temperatures. Moving away from Mode Coupling Theory, ISTS can be used to continue to explore polymer dynamics. Many polymers easily form glasses without crystallization, and thus are an attractive area to extend ISTS measurements, both for practical purposes and theoretical advancement. The contribution of rotational motions is an important question in the interpretation of light scattering data. Orientational relaxation times can be made in the time domain by conducting Optical Kerr effect (OKE) experiments. OKE experiments have been conducted for salol, and questions arise in the comparison of the OKE 150 relaxation times with ISTS relaxation times. 3 Additional insight may be gained from OKE and ISTS or ISBS experiments on the same sample. Sulfur Monochloride, S2C12 is a glass-forming liquid which is also strongly polarizable. As such, it is an attractive candidate for measuring the temperature dependence of orientational relaxation times extending down to the glass transition. Recent experiments have characterized the Kerr signal in the femtosecond and picosecond regime. 4 A cw probe experiment with electronic detection may be useful for measuring longer relaxation times near and possibly below the glass transition. If ISTS or ISBS signal can also be obtained from S2C1 2, some questions about the relationship between ISTS relaxation times and orientational relaxation times may be answered. The use of phase masks and the presence of a controlled "carrier" beam for heterodyne detection creates additional possibilities for experiments related to ISTS. Experiments that previously did not generate sufficient signal levels for observation with a cw probe may be feasible using heterodyne detection. Samples that absorb very weakly may be studied, either through increasing the ISTS signal by heterodyning, or by ISBS. A variation of ISBS with crossed polarizations of the excitation light can measure shear acoustics. 5 Heterodyne detection may simply be used to reduce the amount of excitation and probe energy. Recently, the applicability of ISTS measurements of the properties of thin films for practical application has been demonstrated. 6 In the traditional sense, however, ISTS measurements in liquids still requires excitation energy only available in laboratory lasers. Incorporating phase masks and heterodyne detection, it may be possible to detect signal generated from weaker powered, miniturized diode pump lasers. A possible extension of this work would be a commercially viable ISTS apparatus for real-time monitoring of polymer manufacturing processes. In general, expansion of the characterization of relaxation dynamics, whether by measuring different properties, or by increasing the range of the experiment, can have significant positive impact on the understanding of the behavior of complex systems. Hopefully, the dialog between theory and experiment will continue, improving the physical understanding of relaxation dynamics in complex materials. References 1 Y. Yang, Ph.D. Thesis, Massachusetts Institute of Technology, 1996. 2 A. Duggal, Ph.D. Thesis, Massachusetts Institute of Technology, 1992. 3 R. Torre, P. Bartolini, and R. M. Pick, Phys. Rev. E 57, 1912 (1998). 4 J. Fourkas, (1998). 5 S. Silence, Ph.D. Thesis, Massachusetts Institute of Technology, 1991. 6 J. Rogers, Ph.D. Thesis, Massachusetts Institute of Technology, 1995. 152 Appendix A Selected Matlab Codes Deconvolution of Antel Impulse Response function data=deconvolveimp(fname,simp,tstep) %function to deconvolve impulse responsefrom data data=readdata(fname); N=length (data (:,1)) data=data (:N-1, :); renorm=max (data (:,2)); trep=data (:, 1); data=zerotime (data) ; % set zero of time time=data (:, 1); signal=data (:,2); signal=zerobeg (signal) ; % zero the baseline signal=signal/max(signal); n=sum(time<0); N=length (time); s=signal (n :N) ; % take only positive time data tdata=time (2) -time (1) ; % calculate sampling interval % resample ratio is the ratio of data sampling interval to impulse % sampling interval R=tdata/tstep % resample the signal to the same sampling intervalas the impulse response s=resample (s, round (R) , 1) ; % This is the time consuming step sconv=[ s ' s imp' ' ; % combining the impulse response and data [x,y]=deconv(sconv,simp); x=x(1:length(x)-l); s=resample (x, 1, round (R) ); % resample data back down to the originalrate s=s*renorm/max (s) ; % multiply data back to originalamplitude newsignal=[0.1*signal(l:n-) ' s']'; % put back the originalt<O signal data=[trep(l:round( 0 .95*length(trep))) newsignal(l:round(0.95*length(newsignal)))]; % cut off the end of the data, because often some noise/ringingis picked up % in the deconvolution process Correction for Hamamatsu roll-off function data=hamcorrect (data) % correct datafrom Hamamatsu detector for low frequency rolloff % decay constant is determinedfrom single exponentialfit to early time % response to quasi-cwpulse % do not trust this datafor times longer than 1 microsecond 1); data=zerotime (data); T=data (:, time=data ( :, 1) ; N=length(time); n=sum(time<=0); signal=data (:, 2); decay=ones(size(signal)); decay(n:N)=exp(-time(n:N)/2.2e-06); signal=signal./decay; data=[T signal]; Combination of two data files with different time spacing function [data]=combine(datal, data2) % take signals of different time spacingand combine them % data should already be turned "right-side-up" % input order is short, long timel=datal (:,1); signall=datal(:,2); signall=zerobeg(signall); time2=data2 (:,1); signal2=data2(:,2); signal2=zerobeg(signal2); N1=length (timel) ; N2=length (time2) ; %findamplitude of the end of the short signal levell=mean(signall(N1-50:N1)); tl=timel (N1-50); t2=timel (N1) ; indexl=sum(time2<tl); index2=sum(time2<t2); %findamplitude of overlappingportion of long signal level2=mean(signal2 (indexl :index2)); signall=signall*level2/levell; %scaletomatch signal=[signall' signal2 (index2+1:N2) ']; %combinesignal time=[timel' time2 (index2+1:N2) ']; %combine time axes data= [time; signal] '; ISTS semi-automated fitting routine. Least-squares fitting code is called as an executable. % scriptto fit ISTS data, semi-automatedwith presetfile names % optionfor range of temperaturesor single temperature % yfit.exe must be in the present directory T=input('Enter the temperature filename (in single quotes) or temperature integer>'); directory=input('Enter the directory containing the combined data>', 's'); directory=strcat(directory,'\'); directory2=input('Enter the destination directory>', 's'); directory2=strcat(directory2,'\'); % only save parameterfiles for multiple temperaturefits if isstr(T) % read in the temperatures as integers temps=readdata(T); % initialize matricesforfit parameters % set up Nx2 matrices with temps in thefirst column A=initialize(temps); ga=A;wa=A;tO=A;gh=A;B=A;gr=A;beta=A;chi=A; continue=l; I=1; while and (I<=length(temps), continue==l) fr=input('Do you want to edit fitin.rise? (no=O;yes=l)'); if fr==l ! notepad.exe fitin.rise end T=int2str(temps(I,1)) fname=strcat(directory,T,'all.dat') outfile=strcat(directory2,T,'fit.dat') fidl=fopen(fname, 'r') fid2=fopen(outfile,'w') fclose('all'); if fidl+fid2==7 eval(['! yfit.exe ' fname ' -s ' outfile ' -f fitin.rise']) [fitdata,A,ga,wa,tO, gh,B,gr,beta,chi]=readparams... (outfile,I,A,ga,wa,tO,gh,B,gr,beta,chi); data=readdata (fname); fitdata=fitdata'; figure (1) zoom on subplot(2,1,1), plot(data(:,l), data(:,2), fitdata(:, ) ,fitdata(:,2)) N=length (data ( :, 1)); n=sum(data (:,1)<=0) nl=sum(data(:,l)<data(N,1)/100); subplot(2,1,2), plot(data(l:nl,l),data(l:nl,2),... fitdata(l:nl,l),fitdata(l:nl,2)) figure (2) zoom on semilogx(data(n:N,1), data(n:N,2), fitdata(n:N,1), fitdata(n:N,2)) continue=input('Keep going? (n=0,y=l)'); end I=I+1; end w=input('Save parameters? (no=0;yes=l) '); if w==0 w=input('Save parameters? (no=0;yes=l) '); end if w==1; n=sum (A( :,3) >0); A=A(1:n,:) ga=ga(1l:n,:); wa=wa(1:n, :); t0=t0(1:n,:); gh=gh(l:n,:); B=B(1:n,:); gr=gr(l:n,:); beta=beta(l:n,:); chi=chi(1:n, :); writeparams(directory2,A,ga,wa,t0,gh,B,gr,beta,chi) DWF=[A(:, 1)';A(:,2)';(B(:,3)./(A(:,3)+B(:,3)))']'; dwffile=strcat(directory2,'dwf.dat'); fid=fopen(dwffile, 'w'); fprintf(fid, '%d %d\n',DWF); fclose(fid); end else if input('Do you want to edit fitin.rise? (no=0;yes=l)')==1 I notepad.exe fitin.rise end T=int2str(T); fname=strcat(directory,T,'all.dat'); data=readdata (fname); outfile=strcat(directory2,T,'fit.dat') eval(['! yfit.exe ' fname ' -s ' outfile ' fitin.rise']) % savefit andparameters,getfitdata fitdata=writeonefit(outfile,directory2,T); figure(1); zoom on plot(data(:,1),data(:,2),fitdata(:,1),fitdata(:,2)) end 156 -f Appendix B Selected References Appendix B contains selected references, organized by subject matter. The references to particular citations are found at the end of each chapter. This Appendix contains a more comprehensive list of applicable sources, including some that are not cited in individual chapters. Impulsive Stimulated Scattering Experiment 1. Yan, Y.-x. and K.A. Nelson, Impulsive stimulated light scattering. I. General Theory of Impulsive Stimulated light scattering. II. Comparisonto frequencydomain light-scatteringspectroscopy. Journal of Chemical Physics, 1987. 87(11): p. 6240. 2. Yang, Y. and K. Nelson, Impulsive Stimulated Light Scatteringfrom Glassforming Liquids: I. Generalizedhydrodynamics approach.Journal of Chemical Physics, 1995. 103(18): p. 7722. Mode Coupling Theory and Hydrodynamics and General Liquid-Glass Transition 3. Angell, C.A., Formationof Glassesfrom Liquids and Biopolymers. Science, 1995. 267: p. 1924. 4. Angell, C.A., et al., eds. Structure and Dynamics of Glasses and Glass Formers. Materials Research Society Symposium Proceedings, ed. M.R. Society. Vol. 455. 1997, Materials Research Society: Pittsburgh, PA. 5. Angell, C.A., Perspective on the Glass Transition.Journal of Physical Chemistry of Solids, 1988. 49(8): p. 863. 6. Bengtzelius, U., W. Gotze, and A. Sjolander, Dynamics of Supercooled Liquids and the Glass Transition.Journal of Physics C: Solid State Physics, 1984. 17: p. 5915. 7. Cummins, H.Z., et al., Relaxationaldynamics in supercooled liquids: experimental tests of the mode coupling theory. Physica A, 1994. 204: p. 169. 8. Ediger, M.D., C.A. Angell, and S.R. Nagel, SupercooledLiquids and Glasses. Journal of Physical Chemistry, 1996. 100(31): p. 13200. 9. Fourkas, J.T., et al., eds. SupercooledLiquids: Advances and Novel Applications. ACS Symposium Series. Vol. 676. 1997, American Chemical Society: Washington, DC. 10. Gotze, W. and L. Sjogren, Relaxation Processesin Supercooled Liquids. Reports on Progress in Physics, 1992. 55: p. 241. 11. Hansen, J.P., D. Levesque, and J. Zinn-Justin, eds. Liquids, Freezing,and the Glass Transition.. Vol. 1. 1991, Elsevier Science Publishing Company: Amsterdam. 503. 12. Kim, B., Stretching, Mode Coupling, and the Glass Transition.Physical Review A, 1992. 46(4): p. 1992. 13. Kim, B. and G.F. Mazenko, Mode Coupling, Universality,and the Glass Transition. Physical Review A, 1992. 45(4): p. 2393. 14. Petry, W. and J. Wuttke, QuasielasticNeutron Scattering in Glass Forming Viscous Liquids. Transport Theory in Statistical Physics, 1995. 24: p. 1075. Experimental Studies of Fragile Glass-Forming Liquids 15. Yang, Y. and K. Nelson, Impulsive stimulatedlight scatteringfrom glass-forming liquids: II. Salol relaxationdynamics, nonergodicityparameter,and testing of mode coupling theory. Journal of Chemical Physics, 1995. 103(18): p. 7732. 16. Yang, Y. and K.A. Nelson, Impulsive stimulated thermal scatteringstudy of alpha relaxation dynamics and the Debye-Wallerfactor anomaly in CKN. Journal of Chemical Physics, 1996. 104(14): p. 5429. 17. Cummins, H.Z., et al., Light scatteringspectroscopy of the liquid-glass transtion: comparison with the idealized and extended mode coupling theory. Physica A, 1993: p. 207. 18. Cummins, H.Z., et al., Light-scatteringspectroscopy of the liquid-glass transition in CaKN03 and in the molecular glass Salol: Extended-mode-coupling-theory analysis. Physical Review E., 1993. 47(6): p. 4223. 19. Muller, L., Nonlinear Spectroscopic Studies of Liquids, Ph. D. Thesis in Chemistry. 1995, University of Texas: Austin, TX. 158 20. Schonhals, A., et al., Anomalies in the scaling of the alpha relaxationstudied by dielectric spectroscopy. Physica A, 1993. 201: p. 263. 21. Torre, R., P. Bartolini, and R.M. Pick, Time-resolved optical Kerr effect in a fragile glass-forming liquid, salol. Physical Review E, 1998. 57(2): p. 1912. Experimental Studies of Glycerol 22. Wuttke, J., W. Petry, and S. Pouget, Structuralrelaxationin viscous glycerol: Coherent neutron scattering.Journal of Chemical Physics, 1996. 105(12): p. 5177. 23. Wuttke, J., et al., Neutron and Light ScatteringStudy of Supercooled Glycerol. Physical Review Letters, 1994. 72(19): p. 3052. 24. Birge, N.O. and S.R. Nagel, Specific Heat Spectroscopy of the Glass Transition. Physical Review Letters, 1985. 54(25): p. 2674. 25. Franosch, T., et al., Evolution of structuralrelaxationspectra of glycerol within the gigahertz band. Physical Review E, 1997. 55(3): p. 3183. 26. Jeong, Y.H., S.R. Nagel, and S. Bhattacharya, Ultrasonicinvestigation of the glass transition in glycerol. Physical Review A, 1986. 34(1): p. 602. 27. Lunkenheimer, P., et al., High-frequency dielectric spectroscopy on glycerol. Europhysics Letters, 1996. 33(8): p. 611. 28. Lunkenheimer, P., et al., Fast Dynamics of Glass-FormingGlycerol Studied by Dielectric Spectroscopy. Physical Review Letters, 1996. 77(2): p. 318. 29. Pimenov, A., P. Lunkenheimer, and A. Loidl, BroadbandDielectricSpectroscopy of Glycerol and CKN. Ferroelectrics, 1996. 176: p. 33. 30. Schonhals, A. and E. Schlosser, Relationshipbetween segmental and chain dynamics in polymer melts as studied by dielectric spectroscopy. Physica Scripta, 1993. t49: p. 236. Experimental Studies of Polypropylene Glycol 31. Duggal, A.R. and K.A. Nelson, Picosecond-microsecondstructuralrelaxation dynamics in poly(propylene glycol): Impulsive stimulated light-scattering experiments. Journal of Chemical Physics, 1991. 94: p. 7677. 159 32. Duggal, A.R. and K.A. Nelson, Resolution of conflicting descriptions of poly(propylene glycol) relaxationdynamics through Impulsive Stimulated Scattering experiments. Polymer Communications, 1991. 32(12): p. 356. 33. Johari, G.P., Dielectric relaxation in the liquid and glassy states of poly(propylene oxide) 4000. Polymer, 1986. 27: p. 866. 34. Jones, D.R. and C.H. Wang, DepolarizedRayleigh scatteringand backbone motion ofpolypropylene glycol. Journal of Chemical Physics, 1976. 65(5): p. 1835. 35. Pathmanathan, K., G.P. Johari, and R.K. Chan, Effect of water on relaxations in the glassy and liquid states ofpoly(propylene oxide) of molecular weight 4000. Polymer, 1986. 27: p. 1907. 36. Pathmanathan, K. and G.P. Johari, The effect of water on dielectricrelaxations in the glassy states ofpoly(propylene oxide) andpropylene glycol. Polymer, 1988. 29: p. 303. 37. Sidebottom, D.L., et al., Observation of Scaling Behavior in the Liquid-Glass TransitionRange from Dynamic Light Scattering in Poly(propylene glycol). Physical Review Letters, 1992. 68(24): p. 3587. 38. Baur, M.E. and W.H. Stockmayer, DielectricRelaxation in Liquid Polypropylene Oxides. Journal of Chemical Physics, 1965. 43(12): p. 4319. 39. Bergman, R., et al., Dynamics aroundthe liquid-glasstransition in poly(propylene glycol) investigatedby wide-frequency-range light-scattering techniques. Physical Review B, 1997. 56(18): p. 11619. 40. Borjesson, L., J.R. Stevens, and L.M. Torell, Brillouin scatteringstudies of structuralrelaxationin polypropylene glycol. Polymer, 1987. 28: p. 1803. 41. Huang, Y.Y. and C.H. Wang, Brillouin,Rayleigh, and depolarizedRayleigh scatteringstudies ofpoly(propylene glycol). I Journal of Chemical Physics, 1975. 62(1): p. 120. 42. Wang, C.H. and Y.Y. Huang, Brillouin-Rayleighscatteringstudies of poly(propylene glycol). III. Journal of Chemical Physics, 1976. 64(12): p. 4847. 43. Wang, C.H., et al., Laser light beatingspectroscopicstudies of dynamics in bulk polymers: poly(propylene glycol). Macromolecules, 1981. 14: p. 1363. 160