Structure-property relationships of modified polyesters ... · 2.2.2 NMR relaxometry for polymer...
Transcript of Structure-property relationships of modified polyesters ... · 2.2.2 NMR relaxometry for polymer...
Structure-property relationships of modified polyesters probedby solid-state NMR and FTIR spectroscopyCitation for published version (APA):Ziari, M. (2009). Structure-property relationships of modified polyesters probed by solid-state NMR and FTIRspectroscopy. Eindhoven: Technische Universiteit Eindhoven. https://doi.org/10.6100/IR643420
DOI:10.6100/IR643420
Document status and date:Published: 01/01/2009
Document Version:Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)
Please check the document version of this publication:
• A submitted manuscript is the version of the article upon submission and before peer-review. There can beimportant differences between the submitted version and the official published version of record. Peopleinterested in the research are advised to contact the author for the final version of the publication, or visit theDOI to the publisher's website.• The final author version and the galley proof are versions of the publication after peer review.• The final published version features the final layout of the paper including the volume, issue and pagenumbers.Link to publication
General rightsCopyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright ownersand it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
• Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal.
If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, pleasefollow below link for the End User Agreement:www.tue.nl/taverne
Take down policyIf you believe that this document breaches copyright please contact us at:[email protected] details and we will investigate your claim.
Download date: 04. Jun. 2020
Structure-property relationships of modified polyesters probed by solid-state NMR and FTIR
spectroscopy
PROEFSCHRIFT
ter verkrijging van de graad van doctor aan de Technische Universiteit Eindhoven, op gezag van de
rector magnificus, prof. dr. ir. C. J. van Duijn, voor een commissie aangewezen
door het College voor Promoties in het openbaar te verdedigen op woensdag 24 juni 2009 om 16.00 uur
door
Maya Ziari
geboren te Algiers, Algerije
Dit proefschrift is goedgekeurd door de promotoren: prof.dr. J. Gelan en prof.dr. P. J. Lemstra Copromotor: prof.dr. P. Adriaensens
A catalogue record is available from the Eindhoven University of Technology Library
ISBN: 978-90-386-1896-8
Copyright © 2009 by Maya Ziari
Cover design : Valentine Kreykamp
The Dutch Polymer Institute (DPI) financially supported the research described in this thesis, project
337.
An electronic version of this thesis is downloadable in PDF-format from the website of the Eindhoven
University of Technology (http://www.tue.nl/bib).
In Memory of Laurus, Manola, Farah
“The Science of Today is the Technology of Tomorrow”
To my parents
Table of Contents
v
Table of Contents
Summary ix
Chapter 1. Introduction and scope
1.1 General introduction 1
1.2 Modified polyesters 3
1.3 Characterization of polyesters 5
1.4 Scope of the thesis 7
1.5 Outline of the thesis 8
1.6 References 10
Chapter 2. Basics of NMR relaxation for polymer structure investigation
2.1 General approach 11
2.1.1 Definition and physics basis 11
2.1.2 Pulsed NMR and bulk magnetization 13
2.1.3 Laboratory frame and rotating frame 15
2.2 NMR relaxometry 17
2.2.1 Spin relaxation times 17
2.2.2 NMR relaxometry for polymer characterization 21
2.3 References 30
Chapter 3. Morphology and differences in modified polyesters probed by solid-state NMR
3.1 Introduction 34
3.2 Experimental section 35
3.2.1 Sample preparation 35
3.2.2 NMR experiments 37
3.2.3 NMR pulse sequences 37
3.2.4 NMR fitting procedure 39
3.2.5 Software 39
3.2.6 Modulated DSC 39
Table of Contents
vi
3.3 Results and discussion 40
3.3.1 13C-CP MAS experiments 40
3.3.2 1H T1 and T1 relaxation experiments 43
3.3.3 Quantification of each phase 50
3.3.4 Domains sizes 53
3.4 Conclusion 55
3.5 References 57
Chapter 4. Phase composition of modified polyesters probed by solid-state 1H wideline NMR of spin-spin relaxation
4.1 Introduction 60
4.2 Experimental section 62
4.2.1 Sample preparation 62
4.2.2 NMR experiments 62
4.3 Results and discussion 65
4.3.1 T2 spin-spin relaxation (HEPS) 65
4.3.2 Analysis and relaxation measurements for sample (BD96DI04) 72
4.3.3 Analysis and relaxation measurements for sample (BD54DI46) 73
4.4 Conclusion 78
4.5 References 79
Chapter 5. Heterogeneous chain dynamics in polyester network probed by 1H static spin-spin relaxation NMR experiments
5.1 Introduction 82
5.2 Experimental section 84
5.2.1 Sample preparation 84
5.2.2 NMR experiments 84
5.2.3 Relaxation background 85
5.2.4 Fitting procedure 87
5.2.5 MDSC experiments 89
5.2.6 DMTA experiments 90
5.2.7 DES experiments 90
5.3 Results and discussion 90
Table of Contents
vii
5.3.1 Network heterogeneity and temperature effect on T2 measurement 90
5.3.2 Quantification of rigid and mobile components 102
5.4 Conclusion 107
5.5 References 108
Chapter 6. A FTIR study on the solid-state copolymerization of bis(2-hydroxyethyl)terephthalate and poly(butylene terephthalate) and the resulting copolymers
6.1 Introduction 110
6.2 Experimental section 111
6.3 Theory 112
6.4 Results and discussion 113
6.4.1 Assignment of the absorption bands of the pure components 113
6.4.2 Kinetics of the SSP reaction studied by FTIR spectroscopy 114
6.4.3 Temperature dependent experiments 118
6.5 Conclusion 121
6.6 References 123
Appendix 1. Effect of initial estimates and constrains selection in Multivariate Curve Resolution – Alternating Least Squares. Application to low-resolution NMR data 125
Appendix 2. Combining linear and non-linear regression algorithms as an alternative for multivariate curve resolution problems of low selectivity 149
Appendix 3. Calculation of the (mean) number of monomer units between two cross-linked junctions 161
Acknowledgements 165
Curriculum Vitae 169
Summary
ix
Structure-property relationships of modified polyesters probed by solid-state NMR and FTIR spectroscopy
Summary
In polymer science, a better understanding of the relationship between microscopic
structures and macroscopic properties is essential for an intelligent design of new and/or
improved materials. The objective of this thesis is to contribute to the analysis of
structure-property relationships through the use of non-destructive state-of-the-art NMR
and IR analytical methods in order to characterize polyester co-polymers and chemically
cross-linked unsaturated polyesters.
One of the effects of NMR radio-frequency pulses is to cause the spin populations to
deviate from their equilibrium population. The rate at which the spin system returns to
equilibrium informs about the molecular dynamics and the inter-nuclear distances.
Molecular motions induce nuclear relaxation, since atomic fluctuations modify the
magnetic field experienced by a nuclei and its environment. These fluctuations can
occur in a broad range of frequency and depend directly on the chemical structure and
the morphology. There are different relaxation times that can be measured depending on
the scale of the investigated motions. Fast molecular motions can be investigated by
measuring the so called spin-lattice relaxation time (T1), or spin-lattice relaxation in the
rotating frame (T1) while measuring the spin-spin relaxation time (T2) can be used to
look at slower molecular dynamics.
These different techniques of NMR relaxometry and spectroscopy are exploited in this
thesis to study the molecular dynamics and the morphology of three multi-phase
polymeric systems.
The two first multi-phase polymers have been synthesized by adding two co-monomers,
resp. a diol (2,2-bis[4-(2-hydroxyethoxy)phenyl]propane – Dianol 220®) and a
terephthalate (bis(2-hydroxyethyl)terephthalate – BHET), to poly(butylene
terephthalate) (PBT). Because of its high crystallization rate, PBT is employed in a
broad range of applications, which includes injection-molding grades, films, fibers and
adhesives. In order to extend further its performance, PBT is commonly modified so as
to increase its Tg and improve its dimensional stability, impact strength and
Summary
x
compatibility with other polymers, while retaining its high crystallization rate. Among
using various modification methods, which include copolymerization in the melt,
M.A.G. Janssen [PhD. Thesis, 2005, TU/e] has developed a solid-state
(co)polymerization (SSP) technique to prepare copolymers from polycondensation of
PBT with diols and terephthalates comonomers. In this study, state-of-the-art NMR and
IR analytical methods are exploited, to confirm that diols and terephthalates are
incorporated into the amorphous phase of the PBT by SSP without modification of its
crystalline phase, and more generally to better understand the relationships among
molecular motions, polymer morphology, and macroscopic properties of the multi-phase
PBT systems.
For that purpose, static 1H T1ρ solid-state NMR relaxation experiments have been
performed in combination with cross-polarization magic angle NMR (13C CP-MAS) and
static 1H T2-relaxation measurements on the PBT-Dianol copolymer. Multi-component
1D- and 2D-analysis methods of the proton relaxation behavior showed the presence of
‘’three phases’’ corresponding respectively to the crystalline phase, the rigid-amorphous
phase and mobile-amorphous phase of the polymer systems. Copolymers with different
PBT / Dianol ratios were analyzed: lower Dianol content showed that the Dianol
monomer is exclusively incorporated in the amorphous fractions of the copolymer
system, whereas part of the PBT crystalline phase is also converted to amorphous
copolymer phase at higher Dianol content. The fractions of protons associated to the
different relaxation times, characteristic of each phase (crystalline, rigid-amorphous and
mobile-amorphous), have been determined by using a novel multi-variable analysis
method, based on an orthogonal-projection approach (OPA): correlating line-shape
information to relaxation decay has been shown to be very useful to analyze relaxation
data and better understanding of the molecular dynamics and morphology of multi-
phase polyester systems.
Static 1H T2-relaxation measurements (Solid-Echo and Hahn-Echo) were used to better
estimate the crystallinity in the various samples. As expected, the crystallinity decreased
with increased Dianol concentration. Additional cross-polarization 13C CP-MAS NMR
experiments further confirmed that the Dianol is incorporated into both amorphous
fractions (rigid and mobile). During this study, it was clearly demonstrated that, if a too
high concentration of Dianol is incorporated via SSP, the crystallinity in the sample
decreases drastically as compared to pure PBT. In order to obtain enhanced macroscopic
Summary
xi
properties as compared to pure PBT, the modified polyesters should not exceed a certain
concentration of Dianol. Only then, the glass transition temperature will increase while
the crystallinity and high crystallization rate of pure PBT will remain.
Based on the spin-diffusion phenomenon, 1H T1 and 1H T1 relaxation measurements
have been used to estimate lower and upper limits of the phase domain sizes. The
estimated values between 3 and 80 nm for the rigid phase correlate well with
corresponding SAXS data between 5 and 50 nm.
As for the second PBT-BHET copolymer system, the first objective was to follow the
kinetics of incorporation under isothermal conditions. In that way, the development of
the chemical microstructure during the SSP reaction could be examined and the
miscibility of the BHET and the PBT could also be investigated as a function of
different PBT/BHET ratios. The results show that competition patterns take place
depending on the ratio of PBT/BHET: the structures include annealing of PBT,
incorporation of BHET in amorphous PBT and BHET homo-polymerization. At low
BHET concentrations, BHET is incorporated in PBT forming a non-random copolymer,
while for high BHET concentrations a separate PET phase is found and the PET homo-
polymers have sufficiently large chain lengths to crystallize.
An unsaturated polyester resin (UPR-Palatal intermediate) cross-linked with styrene,
was the third system analyzed. The purpose was to evaluate the use of solid-state NMR
characterization in order to extract micro-structural and molecular dynamics information
of complex chemical cross-linked polymer networks. 1H-NMR T2 relaxation
experiments have been performed to investigate the mobility and heterogeneity in the
network. The results of 1H spin-spin NMR relaxation show a strong heterogeneous
dynamic behavior. Also here three types of 1H mobilities were observed above Tg but
the different molecular mobility parts in the polymer network are now correlated to the
inter cross-link chains, the dangling chains ends and the sol fraction. The average
molecular weight between cross-links was investigated and estimated to be only a few
polystyrene monomer units. The data obtained via other characterization techniques,
such DMTA, DES and M-DSC, confirmed these results.
Chapter 1
1
Chapter 1
Introduction and Scope
1.1. General introduction and historical perspectives
Polyesters represent a class of polymers which contain the ester functional group in their
main chain. Although polyesters do exist in nature (natural polyesters have been known
since around 1830), the term polyester generally refers to the large family of synthetic
polyesters (resins), which includes polycarbonates and polyethylene terephthalates
(PET) [1,2].
Polyesters are widely used materials with diverse applications (fibers, plastics,
coatings). They are strong, colorfast, and resistant to corrosion and chemical attack. In
general they have extremely good mechanical properties and are heat resistant.
The first polyester fibers were synthesized by Carothers, also the inventor of nylon
fibers. His work was resumed by a group of British scientists, J.R. Whinfield, J.T.
Dickson, W.K. Birtwhistle, and C.G. Ritchie, who successfully created the first
polyester fiber called Terylene in 1941 [3]. A few years later, Dupont came up with an
other polyester fiber, the Dacron®. Today, polyester fibers are widely used in the textile
industry. For that purpose, they are often spun together with fibers of cotton, producing
a cloth with some of the better properties of each.
Polyesters are polymers made by a polycondensation reaction in which the linkage
between the molecules occurs through the formation of ester groups.
We can distinguish three main categories of polyesters, depending on their properties
and type of applications:
1. The thermoplastic engineering types, such as poly(ethylene terephthalate) (PET),
and the poly(butylene terephthalate) (PBT);
2. The thermoset composites such as unsaturated polyester resins;
3. The aromatic liquid-crystalline polyesters.
Chapter 1
2
Poly(ethylene terephthalate) PET is the most widely used linear polyester. It is prepared
from terephthalic acid and ethylene glycol (35 million tones/annum). Such polymer can
be prepared in bulk or in solution using an excess of ethylene glycol to increase the
esterification rate. This reaction leads first to a low-molecular-weight hydroxyl-
terminated polyester, which is then transesterified with removal of excess glycol to
attain high molecular weight. One should note that terephthalic acid is quite insoluble in
common solvents and sublimes at 300 oC. For this reason, dimethyl terephthalate
(DMT) has been used more frequently to make PET. Another process based on ethylene
oxide instead of ethylene glycol can also be used. Ethylene oxide reacts rapidly with
terephthalic acid to form an intermediate which will undergoes alcoholysis to finally
form the polyester.
PET is a slowly crystallizing polymer and is therefore usually in the amorphous state
and hence transparent, but can become translucent when it is in its semi-crystalline state.
With excellent barrier (CO2 permeation) and mechanical properties, PET is becoming
the polymer of choice for food and beverage packaging applications, even replacing
traditional materials such as metals or glass. The addition of anti-nucleating agents
permits the injection molding of PET and broadens its application range. However, for
an injection molding type of application, polybutylene terephthalate (PBT) is preferably
used. PBT as well is a semi-crystalline polyester, produced at an annual rate of about
400 thousand tons. It is characterized by a high crystallization rate, and good thermal,
mechanical and electrical properties. Such a polyester is polymerized following a two-
stage process. First, bis(4-hydroxybutyl)-terephthalate (BHBT) is formed by trans-
esterification of DMT with 1,4 butanediol, then BHBT further reacts by
polycondensation and leads to PBT. In comparison with PET, PBT has a slightly lower
strength and rigidity, a slightly better impact resistance, and a slightly lower glass
transition temperature. PBT is used for door and window hardware, automobile luggage
racks and body panels, headlight reflectors, and fiber optic cables. It is being used to
replace PVC for cable sheathing, due to a combination of environmental pressures and
outstanding electrical properties across a wide range of temperatures. Last but not least,
PBT is also replacing metals (in conjunction with polycarbonates, PC, as an alloy/blend)
for automotive body panels.
Chapter 1
3
Generally the use of PET and PBT blends broadens the range of application of these
thermoplastic engineering-grade polyesters.
Unsaturated polyester resins (UPR) are the third-largest class of thermoset molding
resins. They constitute the most common polymers used in conjunction with glass fiber
reinforcing. Typical applications are in fiberglass-reinforced shower stalls, boat hulls,
wind turbine blades, construction panels and autobody parts. UPR are prepared in a step
polymerization process from a glycol, typically 1,2 propylene glycol together with both
a saturated and an unsaturated dicarboxylic acid. The unsaturated acid provides sites for
cross-linking along the backbone, while the saturated acids effectively limit the number
of cross-link points. The effect of limiting theses sites is to reduce the brittleness of the
cured resin. They are numerous possible monomers that can be used in the backbone of
the polyester prepolymers. Typical monomer includes 1,2-propanediol, with maleic acid
(usually added as an anhydride) and phtalic acid (again usually added as an anhydride).
Cross-linking is initiated through an exothermic reaction involving organic peroxides,
such as methyl ethyl ketone peroxide or benzoyl peroxide [4].
A relatively new class of polyesters is the so-called liquid-crystalline polyesters which
possess excellent flow properties, but their use is still limited. Hence this class of
polyesters will not be discussed in the thesis [5].
1.2. Modified polyesters
In general, polyesters in bulk do not fulfill all requirements of industrial applications.
Ways to overcome this is to modify the bulk properties by melt-reactive blending, for
instance with other polymers, by copolymerization in the melt or in the solid-state or by
cross-linking the material [6]. In the first case, miscibility of the blend is essential. This
implies keeping two homoploymers in the molten state in the presence of a catalyst to
enhance the trans-esterification reactions [7]. In that way, a fully random copolymer is
obtained [8,9].
Several studies about modification of polyesters can be found in literature, for example,
about the modification of poly(butadiene terephtalate) (PBT) using reactive blending
[9,10].
Chapter 1
4
Generally, polyester is modified through the so-called melt polymerisation (MP). The
principle of MP is based on mixing a diol monomer with a homopolymer in the
presence of a catalyst [11]. The diol monomer has to fulfill some conditions, such as
having a melting temperature (Tm) below that of the homopolyester and being thermally
stable. During the incorporation of the diol monomer, the original diol of the
homopolyester should evaporate. The end product is a modified polyester with diol
monomer and it is a fully random copolymer. The two previous methods of modification
of polyesters are mainly suited for amorphous polymers. For semi-crystalline polyesters,
the situation gets more complicated. In fact, the resulting copolymers have a shorter and
more irregular crystalline structure then the original homopolyester. This implies a
decrease in the melting temperature, crystallinity, and crystallization rate, so that the
final macroscopic properties will be affected.
During the last decade, studies have been reported on ways to modify semi-crystalline
polyesters in a controlled way, so that the crystallization behavior of the original
polyester is retained, while the mechanical properties are enhanced. One that caught our
attention is the so-called solid-state polymerization (SSP) [12,13,14]. SSP is generally
used to increase the molecular weight of semi-crystalline condensates [15]. The
principle of SSP is based on performing a reaction at a temperature well above the glass
transition temperature (Tg) of the original homopolyester and just below its melting
temperature (Tm). In that way the polymer chains in the amorphous phase are highly
mobile and the transesterification reaction with the other polycondensate can occur.
These reactions are generally of the type of outer-outer transesterification reactions [14].
Another method to modify polyesters is through cross-linking. Cross-linking can be
introduced into an assembly of polymer molecules either as the polymerization takes
place or as a separate step after the initial macromolecule has been formed.
Covalent chemical bonds that occur between macromolecules are known as chemical
cross-links. Their presence and their density have a major influence on both chemical
and mechanical properties of the material in which they occur. The presence of cross-
links between macromolecules influences the way in which these materials respond to
heat. As mentioned earlier, uncross-linked polymers will generally melt and flow at
sufficiently high temperatures; they are usually thermoplastics. By contrast, cross-linked
polymers cannot melt because of the constraints of molecular motions introduced by the
Chapter 1
5
cross-links. Instead, at high enough temperature, they will undergo irreversible
deformation. Such polymers will also not dissolve, as the solvatation of chain segments
cannot overcome the effect of the covalent bonds between the macromolecules.
However, depending on the cross-link density, such material may admit significant
amount of solvent, becoming softer and swollen. Such swelling is reversible.
The mechanical properties of the cross-linked polymers depend on the degree of cross-
linking. Heavily cross-linked polymers tend to be very brittle. Indeed, heavily cross-
linked materials contain a dense three-dimensional network of covalent bonds that
implies very little degree of freedom for molecular motion. Hence there is no
mechanism available to allow the material to take up the stress.
Certain commercially important cross-linking reactions are carried out with unsaturated
polymers. One of the widely used ones involves unsaturated polyesters. As already
mentioned in the first part of this chapter, such polyesters are obtained using
bifunctional acids that contain double bounds. In most cases the cross-linking step
consists of adding a solvent monomer (usually styrene) to the unsaturated polyester,
followed by curing the system. The cross-linking reaction is based on a typical radical
polymerization reaction, initiated by the presence of a peroxide, activated by a redox
reaction with cobalt salts (or thermally). This kind of reaction is called curing. In
general, during the curing reaction, the physical properties of the polyester are changing
such as the shear, the torsion modulus, the hardness, the dielectric constant and the
viscosity. The idea is to generate a three-dimensional network with good mechanical
properties and strength impact [16-20].
1.3. Characterization of polyesters
We can divide the modified polyesters in two classes: on the one hand the thermoplastic
materials such as modified polyesters obtained via reactive blending or via MP, or SSP,
and on the other hand the thermoset cross-linked polyesters, which consist of a rigid
network. Depending on the class of polyesters, different characterization techniques are
required. In general, modified polyesters such as modified PBT or PET, can be
characterized with liquid-state Nuclear Magnetic Resonance Spectroscopy (NMR), Size
Exclusion Chromatography (SEC), High-Performance Liquid-Chromatography (HPLC),
Chapter 1
6
Differential Scanning Calorimetry (DSC), and Small-Angle X-ray Scattering (SAXS). 1H liquid-state NMR and solid-state NMR could be used to study the chemical
microstructure [12,14]. Quantitative 13C NMR can also be used to study the sequence
distribution in order to determine the degree of randomness (R) of the copolymer. DSC
could be used to investigate the thermal properties of the material as well as the
morphology. Diffraction techniques can also be used for that purpose. The average
molecular weight (Mw) of the polymer can be determined by chromatographic
techniques. Characterization of polymer networks is a much more complex task. This
has captivated the attention of many polymer scientists, especially during the last two
decades [15,16,18-26]. To be able to improve and optimize the macroscopic properties
of cross-linked resins, it is crucial to develop a thorough understanding at the
microscopic level, including features such as the cross-link density of a network, the
number of dangling chains, loops, and eventual polymer chains unattached to the
network (commonly called the sol fraction). The chemical structure (topology) of cross-
linked networks is extremely complicated [19, 20]. During the formation of a network,
many different structures may be formed, resulting in a very heterogeneous network.
Different initiation, propagation, termination, and combination reactions may occur, as
well as side reactions with other components present in the curing resin system. Many
parameters affect the curing process, including the temperature, the solvent, the
chemical and physical properties of the (pre)-polymers, the additives and the catalysts
present. The final network structure with all its variations, heterogeneities and defects
determines the properties of the network [21].
Not many analytical techniques can be applied to study the chemical structure of
polyester networks. Fourier-Transform InfraRed (FTIR) and High-Resolution Solid-
State Nuclear-Magnetic Resonance (NMR) spectroscopy will yield information on the
chemical composition of the network, such as the number of residual double bonds.
However, this information pertains to an average structure, rather than to the actual
chemical moieties present. In other words, the heterogeneity of the network is not
reflected. Also, it is very difficult to use such spectroscopic methods to obtain accurate
quantitative data on networks.
The morphology of polymeric networks can be studied with X-ray and neutron-
scattering techniques, or with spectroscopic methods. The physical and mechanical
Chapter 1
7
properties of the networks can be studied using thermal and mechanical techniques,
such as Differential Scanning Calorimetry (DSC), Dielectric Spectroscopy (DES) and
Dynamic Thermal Mechanical Analysis (DMTA). Also, solid-state NMR relaxometry
(mainly 1H T2-relaxation) has proven to be a valuable technique for studying various
types of polymeric networks; these include, for example, cross-linked epoxy resins [22,
23], cross-linked poly(styrenes) [24], and cured EPDM [25], as well as network resins
and unsaturated polyesters [26].
1.4. Scope of the thesis
The objective of this project is to study and investigate structure-property relationship of
modified polyester samples using solid-state NMR spectroscopy and infrared
spectroscopy (FTIR) as the main characterization techniques. Three types of modified
polyesters have been studied: two types of modified poly(butylene terephthalates) (PBT)
with diol monomers and an unsaturated polyester network.
We studied the microstructure of modified PBT. Two types of copolymers have been
studied; one copolymer resulting from the incorporation of 2,2-bis[4-(2-
hydroxyethoxy)phenyl]propane monomers also known as Dianol 220 and the other
copolymer resulting from the incorporation of bis(2-hydroxyethyl)terephthalate
monomer also called BHET. Both monomers have been incorporated via solid-state
polymerization (SSP). For the first copolymer a three phases system morphology was to
be demonstrated. For that purpose, 1H T1 and 1H T1ρ experiments were performed in
combination with cross-polarization NMR and 1H T2-relaxation measurements.
Mixtures with different PBT / Dianol ratios were analyzed.
On the second copolymer system, the incorporation of the BHET monomer in PBT via
SSP was investigated. The idea was to follow the kinetics of incorporation under
isothermal conditions. In that way, the development of the chemical microstructure
during the SSP reaction could be examined, and the miscibility of the BHET and the
PBT could also be investigated. Samples with different PBT/BHET ratios were also
analyzed. The kinetics of the SSP reaction was followed using infrared spectroscopy.
The microstructure and thermal behavior of the different copolymer samples were
Chapter 1
8
analyzed using two-dimensional correlation infrared spectroscopy. The presence of PET
homopolymer (blended material) was to be demonstrated.
Furthermore we studied the molecular dynamics of a cross-linked polyester network. It
is well known that the characterization of such chemically cross-link 3D-network is a
very challenging task. 1H-NMR T2 relaxation experiments have been performed to
investigate the mobility and the heterogeneity of the network. Solid-echo and Hahn-
echo pulse sequences have been applied. The data were fitted using combinations of
Gaussian and exponential decays. A 1 dimensional (1D) and a (2 D) approach have been
used for the analyses.
1.5. Outline of the thesis
In Chapter 2, the basic principles of solid-state NMR are reviewed, based on the
classical physics approach. Also NMR relaxation mechanisms are described and related
to the molecular dynamics and the morphology of the polymeric materials.
In Chapter 3, solid-state NMR spectroscopy and relaxometry have been used to
determine the morphology of the final products of the incorporation of a Dianol
monomer into poly(butylene terephthalate) (PBT). A series of these modified polyesters
has been investigated with different dianol / PBT ratios. Solid-state NMR of 1H spin
diffusion measurements, spin-lattice relaxation times T1 and T1 (in the rotating frame)
were used to investigate the morphology in combination with T1 filtered 1H-13C cross-
polarization experiments. The results have been compared to these obtained through
SAXS and TMDSC techniques.
In Chapter 4, static solid-state 1H spin-spin or transverse relaxation (T2) experiments
were used to investigate the rigid fractions of the modified PBT samples. The results
were fitted with a multi-exponential decay model, or a combination of Gaussian and
exponential decays. Solid-echo as Hahn-echo pulse sequences have been applied.
In Chapter 5, heterogeneous chain dynamics in a cured unsaturated polyester resin
(UPR-Palatal intermediate) are investigated using 1H-NMR transverse relaxation
experiments (T2). Different transverse relaxation times (T2) are reported at different
temperatures. T2 measurements were performed on a cross-linked sample in the non-
Chapter 1
9
swollen and the swollen state, at respectively 4.7 and 11.7 Tesla by means of a novel
multi-variable procedure (Appendix 1& 2). Spectral resolution has been implemented
at high field, correlating line-shape information to relaxation decay. The obtained two-
dimensional data sets were analyzed in terms of two decaying components without any
assumption about the underlying line shape. Furthermore, Dynamic Mechanical
Thermal Analysis (DMTA) and Dielectric Spectroscopy (DES) were used to provide
complementary information on the network mobility.
In Chapter 6, the kinetics of incorporation of the bis (2-hydroxyethyl)terephtalate
(BHET) monomer in poly(butylene terephthalate) (PBT) via solid-state polymerization
(SSP) was investigated using infrared spectroscopy. In that way, the development of the
chemical microstructure during the SSP reaction could be examined, and the miscibility
of the BHET and the PBT could also be investigated. Samples with different
PBT/BHET ratios were characterized. The microstructure and thermal behavior of the
different copolymer samples obtained through SSP were analyzed using two-
dimensional correlation infrared spectroscopy. The presence of PET homopolymer
(blended material) was demonstrated.
1.6. References
[1] Van Berkel, R. W. M.; Van Hartingsveldt, E. A. A.; van der Sluijs, C. L. Handbook of Thermoplastics, Marcel Dekker Ed. 1997.
[2] Young, R. J.; Lovell, P. A. Introduction to Polymers, Chapman & Hall 2nd Ed. 1991.
[3] Flory P. J. Principles of Polymer Chemistry, Cornell Press Ithaca 1953.
[4] Coran, A. Y. Encyclopedia of Polymer Engineering, John Willey & Sons 2nd Ed. 1989.
[5] Mucha, M. Colloid &Polymer Science 1991, 269, 1435.
[6] Devaux, J.; Godard, P.; Mercier, J. P.; Touillaux, R.; Dereppe, J. M. J.Polym. Sci., Polym. Phys. 1982, 20, 1881.
[7] Backson, S. C. E.; Kenwright, A. M.; Richards, R. W. Polymer 1995, 36, 1991.
[8] Kim, J. H; Lyoo, W. S.; Ha, W. S. J. Appl. Polym. Sci. 2001, 82, 159.
[9] Marchese, P.; Celli, A; Fiorini, M. J. Polym. Sci., Polym. Phys. 2004, 42, 2821.
[10] Fernandez-Berrediti, M. J.; Iruin, J. J.; Maiza, I. Polymer 1995, 36, 1357.
[11] Berti C.; Colonna, M.; Fiorini, M.; Lorenzetti, C.; Marchese, P. Macromol. Mat. Eng. 2004, 289, 49.
[12] Hait, S. B.; Sivaram, S. Macromol. Chem. Phys. 1998, 199, 2689.
[13] Kimura, M.; Porter .R. S. J.Polym. Sci., Polym. Phys. 1983, 21, 367.
[14] Jansen, M. A. G.; Goossens, J. G. P.; de Wit, G.; Bailly, C.; Koning, C. E. Macromolecules 2005, 38, 2659.
[15] Jansen, M. A. G.; Goossens, J. P. G.; de Wit, G.; Bailly, C.; Koning C. E. Anal. Chim. Acta 2006, 557, 19.
[16] Litvinov, V. M.; Prajna, P. Spectroscopy of Rubbers and Rubbery Materials 2002, 360.
[17] Vilgis, T. A.; Heinrich, G. Coll. Pol. Sci. 1991, 269, 1003.
[18] Harrison, D. P. J.; Yates, W. R.; Johnson, J. F. J. Macromol. Sci. 1985, 25, 481.
[19] Schimmel, K-H.; Heinrich, G. Coll. Pol. Sci. 1992, 242, 1008.
[20] Vilgis, T. A.; Heinrich, G. Macromol. Theory and Sim. 1994, 3, 271.
[21] Woll, R. P. Macromolecule 1993, 26, 1564.
[22] O’Donnell, J. H.; Whittaker A. K. Polymer 1992, 33, 62.
[23] Orza, R. A.; Magusin, P. C. M. M.; Litvinov, V. M.; van Duin, M.; Michels. M. A. J. Macro. Symp. 2007, 40, 8999.
[24] Mohanraj, S.; Ford, W. T. Macromolecules 1985, 18, 351.
[25] Winters, R.; Lugtenburg, J.; Litvinov, V. M. Polymer 2001, 42, 24.
[26] Spyros, A. Journal of Applied Polymer Science 2003, 88, 1881.
10
Chapter 2
11
Chapter 2
Basics of NMR relaxation for polymer structure investigation
2.1. General approach
Nuclear magnetic resonance spectroscopy (NMR) uses the spin resonance phenomenon
to study physical, chemical, and biological properties of matter. As a consequence,
NMR spectroscopy finds applications in several areas of science.
NMR is a very powerful analytical tool to investigate molecular structure-property
relationships of polymer systems [1-8]. NMR can be used, on the one hand, as a
spectroscopic technique to obtain information about packing of the polymer, chemical
structure and conformation of the molecules. On the other hand, NMR can be used as a
relaxometry technique to probe chain dynamics and morphology of polymers [9-12].
This chapter should provide the reader with an understanding of the principles of NMR
from the microscopic, macroscopic and materials (polymer) perspectives. First, the
basic principles of solid-state NMR are reviewed based on the classical approach of
physics, then NMR relaxation mechanisms are described and related to the molecular
dynamics and the morphology of polymeric materials.
2.1.1. Definition and physics basis
Nuclear magnetic resonance is a physical phenomenon of atomic nuclei in a magnetic
field, which can be used to investigate molecular properties of matter [13-17]. Atomic
nuclei of most isotopes, except those with both even proton and neutron numbers,
possess a property called spin. Spin is a quantum mechanical operator I = (Ix, Iy, Iz)
proportional to the angular momentum J = I with Planck’s constant = h/2. I can
have different orientations, but its size is restricted to )1( II where I is an isotope-
dependent spin quantum number.
Chapter 2
12
Nuclei with non-zero spin also have a magnetic moment:
Iμ (2.1)
with the so-called gyromagnetic ratio, which is an isotope dependent property. For
polymer materials the 1H- and 13C-nuclei are the most important. Both have spin
quantum number I = ½ and therefore have two quantum states |> and |> with different
projections <Iz> of I on the magnetic field B0, respectively, m = ½ and m = -½. As
follows from the Zeeman Hamiltonian for a spin interacting with a magnetic field with
size B0:
0BzIH (2.2)
the spin-up state |> has a lower energy E –½ B0 and the spin-down state |> has a
higher energy E ½ B0 . Transitions between the and state can be induced by
radio-frequency (rf) radiation with the proper angular frequency:
/0 EE B0 (2.3)
This is the basis of nuclear magnetic resonance. Instead of this quantum mechanical
picture of NMR, a convenient semi-classical picture is nowadays used for NMR. In
thermodynamic equilibrium, an ensemble of nuclear spins in a magnetic field B0 has a
net alignment of the overall nuclear magnetic moment along B0. A short rf pulse at their
resonance frequency o will tilt this overall magnetization away from B0. After this
pulse, the magnetization will precess around the magnetic field axis with frequency o =
B0 according to the equation of motion for angular momentum J under the influence of
a torque x B0 which tries to align along B0:
0BμJ
dt
d (2.4)
or, after substitution J = I and = I = J, in terms of the magnetic moment :
0Bμμ
dt
d (2.5)
and in terms of the spin operator
0BII
dt
d (2.6)
Chapter 2
13
Figure 2.1.a. Orientation of the magnetic moment () vectors in a static magnetic field B0 (left). Figure 2.1.b. Energy scheme of the nuclei (I=1/2) (right).
According to Planck’s law, the energy necessary to reverse the orientation of the
magnetic moment or to produce a transition between the energy levels equals the
difference in the allowed energy levels. The frequency of the radiofrequency (rf)
irradiation ( 0 ) needed to cause such a transition is proportional to the Larmor
frequency (0). The difference in energy is given by the following equation:
00
2
1
2
1 hν.ωEEΔΕ
(2.7)
2.1.2. Pulsed NMR and bulk magnetization
It is rather cumbersome to describe NMR on a microscopic scale. A macroscopic picture
is more convenient. In an NMR experiment, a large number of nuclei are observed (in
the order of 1023). The vector sum of the magnetization vectors from all the individual
nuclei results in a net or bulk magnetization or M0 that aligns at equilibrium along the
direction of the field B0 along z-axis (Figure 2.2). In order to describe pulsed NMR it is
necessary to talk in terms of the net or bulk magnetization.
As already mentioned above, when a group of spins are placed into a magnetic field B0,
the spin aligns in one of the two possible orientations.
At room temperature, the number of spins in the lower energy level, N, slightly
outnumbers the number in the upper level, N. Boltzmann statistics tells us that:
Chapter 2
14
TE/k -
α
β beN
N (2.8)
E is the energy difference between the spin states; k is the Boltzman's constant,
1.3805x10-23 J/K, and T is the absolute temperature. One could observe that the ratio N
/ N decreases with decreasing temperature resulting in an increased sensitivity.
The value of the net magnetization at equilibrium is given by the Curie’s law as a
function of the static magnetic field B0, the temperature of the system T, the number of
spins in the sample N, the spin I and the gyro-magnetic ratio (equation 2.9):
0B..T.k
).I(I.N.γM
B3
122
0
(2.9)
In the NMR experiment the frequency of the photon is in the radio frequency (rf) range.
In current 1H NMR spectroscopy, 0 ranges between 60 and 1000 MHz.
In pulse NMR the signal is proportional to the population difference between the states.
NMR is in principal a rather insensitive spectroscopic technique because of the small
population differences at the Boltzmann equilibrium state with the small h quanta.
Thanks to the state- of- the- art electronics, very small population differences can be
detected in a very efficient way.
Figure 2.2. Precession of the magnetic moment around a double cone (left). Distributions of all the magnetic moments (), on upper () and lower energy level () with N N. A macroscopic resultant called net magnetization is observed (or bulk magnetization M0 (right)) .
Chapter 2
15
2.1.3. Laboratory frame and rotating frame
In an NMR experiment, transitions are induced between two spin states by irradiating
the nuclei with a superimposed radiofrequency (rf) pulse (also called B1 field) applied
perpendicularly to the original field B0. The net magnetization vector M0 is tipped away
from the z- axis, describing a complex spiral motion going up and down along the z
axis. The circular motion is in a MHz frequency-range (rf), while the nutation is in the
kHz range.
This is the situation in the laboratory frame (xyz). Using a rotating frame (x’y’z), that
rotates with frequency rf, does fix the B1 vector along the x’ axis. In that case, M0
describes a simple circular motion in the z-y’ plane.
To understand how a small magnetic field B1 is able to rotate the magnetization away
from the z-axis even in the presence of a strong magnetic field B0, it is necessary to look
at the motion in a rotating frame of reference R’(x’y’z).
In the laboratory frame R the atomic magnetization vectors precess about B0 at a
constant angle and at an angular frequency 0 . When a rf pulse is applied along the x-
axis (B1) the magnetization will precess around an effective magnetic field Beff at an
angular frequency eff . The rf field B1 oscillates about B0 at an angular frequency rf .
In the rotating frame R’ the magnetization is seen to precess around the static magnetic
field at an apparent frequency ( 0 - rf ).
According to equation (2.1), the static magnetic field will not be B0 anymore but a
reduced magnetic field: ∆B = ( 0 - rf )/. When o = rf, the apparent Larmor
frequency will be 0 and the static magnetic field will vanish as described in Figure 2.3.
In the rotating frame B0 seems to shrink and B1 becomes dominant. The rotation angle
in the plane z-y’ depends on the time the transmitter field B1 is on (the pulse width) p
and its magnitude:
1B p (2.10)
Chapter 2
16
z z
∆B Ω Beff ωeff
x xB1 ω1
Figure 2.3. In the rotating frame, the effective field Beff is the vectorial sum of the reduced field B and the B1 field (left). We can also express the Beff in term of frequencies (right). is defined as the angle between B and Beff.
As observed in Figure 2.4 a 90 pulse rotates the equilibrium magnetization down to the
y’- axis. Stopping the transmitter B1 after a 90° pulse results in a maximal transverse
magnetization rotating about the z- axis and will induce a current in a coil of wire in the
transversal plane. The receiver will detect a current in the x’y’-plane which gives a sine
wave as a function of time. This wave however will decay with a time constant T2* (see
chapter 2.2.1.2.) due to dephasing of the spins. This signal is the free induction decay
(FID). The FID is converted into the frequency domain spectrum by a Fourier
Transformation (FT) (Figure 2.5).
z
y’
x’
B1
My’
z
y’
x’
B1
My’
Figure 2.4. Direction of the net magnetization vector M0 after a 90 x’. A 180 pulse rotates the equilibrium magnetization down to along the z axis.
Chapter 2
17
Figure 2.5. Decay of the transverse magnetization M, in the time (left) and frequency domain after Fourier-transformation, FT, (right).
2.2. NMR relaxometry
2.2.1. Spin Relaxation times
2.2.1.1. Spin-lattice relaxation time (T1)
At equilibrium, the net magnetization vector lies along the direction of the applied
magnetic field B0 and is called the equilibrium magnetization M0. In this configuration,
the z component of magnetization MZ equals M0. MZ is referred to as the longitudinal
magnetization. When a 90°-pulse is applied, the magnetization along B0 disappears and
the population of the two levels becomes the same (or inverted after a 180 pulse) as
seen in Figures 2.6.
The process to return to the Boltzmann equilibrium state along the z-axis is the so-called
spin-lattice relaxation time or longitudinal relaxation time T1.
Inversion recovery (Figure 2.6.b) is mostly used to determine T1 relaxation. In that case,
the magnetization Mz is described by equation (2.11):
(2.11))21( 1/0
Ttz eMM
Chapter 2
18
y’
z
y’
x’
B1
My’
n = n
z
x’
B1
My’
n > n
M-z
z-
(a) (b)
y’
z
y’
x’
B1
My’
n = n
z
x’
B1
My’
n > n
M-z
z-
(a) (b)z
y’
x’
B1
My’
z
y’
x’
B1
My’
n = n
z
x’
B1
My’
n > n
M-z
z-
(a) (b)
Figure 2.6.(a): direction of the net magnetization vector M0 after a 90x’- pulse, so called saturation recovery (left). Figure 2.6.(b): after a 180x’- pulse, so called inversion recovery (right).
Also a 90°-pulse can be used, the so-called saturation recovery (Figures 2.6.a and 2.6.b);
in that case, the return of the Mz magnetization to equilibrium value M0 is described by
the following equation:
(2.12) )1( 1/0
Ttz eMM
The interactions of the magnetization vectors with the changing electromagnetic dipoles
in the lattice result in a transfer of energy from the spin system to the lattice. This spin
lattice relaxation process is an enthalpic process and a first order process that can be
described by the Bloch equation [16] as below:
1
0 )(
T
MM
dt
dM zz (2.13)
As already mentioned, most of T1 relaxation time measurements are performed by
means of the inversion recovery method. Due to the repetitive character of pulse NMR
spectroscopy, T1 is heavily related to its quantitative aspect. Effectively, before applying
a new pulse sequence, it is necessary for Mo to recover its Boltzmann equilibrium.
Therefore, to obtain quantitative results, a delay time of 5 times T1 should be respected
between two successive pulse sequences.
Chapter 2
19
2.2.1.2. Spin-spin relaxation (T2)
Spin-spin relaxation is defined as the decay to zero of the transverse component of M0
(the magnetization My’). When applying a pulse ( pulse), the magnetization M0
along the z- axis is not only flipped away from the direction of B0, but also the
precession movements of the magnetic moments are synchronized (coherence). As in
the case of a oscillatory movement, the magnetic moments precess at the same
frequency and phase. The de-phasing (loss of coherence) starts when B1 = 0. The
precession movements of each spin are no more synchronized, as each spin will
experience a slightly different magnetic field caused by spin-spin interactions. After a
certain time, the projection onto the x’-y’ plane are uniformly distributed around B0 as
seen in Figure 2.7. After a time T2, the transverse components of the magnetizations Mx’
and My’ are reduced to 0.37 M0 (2.15). This time is the so- called spin-spin relaxation
time. What is measured is the lost of coherence between the spins as a function of time.
'90 x
The decay of the x’-y’ magnetization may be written using the Bloch’s equations [16]:
2
''
T
M
dt
dMxx (2.14.1)
2
''
T
M
dt
dMyy (2.14.2)
During spin-spin relaxation the magnetization is transferred from one spin to another
without changing the net energy of the system (flip flop mechanism) that is an entropic
process; no change in population of the energy levels occurs.
The decay of the transverse magnetization My’ is also described by an exponential
equation:
(2.15) 2/0'
Tty eMM
T2 cannot be longer than T1. When the longitudinal magnetization has recovered its
maximal value aligned along the z-axis, the transverse magnetization is 0. The study of
T2 reveals important information about the structure of the matter since T2 determines
the line width of the signal in the frequency spectrum. This line width probes
information about local mobility of the sample as explained in the following section
(2.2.2).
Chapter 2
20
x’ y’ x’ y’
x’ y’ x’ y’
x’ y’
(a) (b)
(c) (d)
(e)
x’ y’ x’ y’
x’ y’ x’ y’
x’ y’
(a) (b)
(c) (d)
(e)
Figure 2.7. Loss of coherence between the spins after a 90 x’ pulse (a, b, c), after a partial dephasing (d) and after complete loss of coherence (e) (My’= 0).
The line width is given by the following relation:
2
2/1.
1
T (2.16)
As the static magnetic field B0 is not exactly homogeneous, field inhomogeneities cause
nuclei (even the ones that are chemically equivalent) to precess with slightly different
Larmor frequencies. Therefore T2* is the spin-spin relaxation time that incorporates field
inhomogeneities:
1/ = 1/T2mat + 1/T2inhom (2.17) 2T
The experimental pulse sequences to determine and T2 are described in Chapter 4. 2T
2.2.1.3. Spin lattice in the rotating frame (T1)
As introduced earlier, a classical pulse will bring the magnetization in the x’y’
plane. If this is immediately followed by a 90 phase-shift of the transmitter, also the B1
field will lie along y’ and will lock the magnetization on the y’-axis as seen in Figure
2.8. The relaxation of the magnetization vector becomes more difficult because of the
competition between B1 and B0. However B0 is a much stronger field, so the
magnetization will relax much more slowly. The time it will take for the spin lock
'90 x
Chapter 2
21
magnetization to relax is the so-called spin-lattice relaxation in the rotating frame (T1).
As for T1, the lattice also influences T1. This is also an enthalpy process. Molecular
motions stimulate this relaxation most efficiently when their frequencies match the
frequencies dictated by the strength of the spin lock field B1 (1 =γ. B1) that is in the
order of tens of kHz. The experimental pulse sequence to determine T1 is described in
Chapter 3.
Figure 2.8 (a, b) 90 x’ pulse and phase shift. (c, d ) spin-lock of the magnetization along y’-axis.
2.2.2. NMR relaxometry for polymer characterization
Besides classical spectroscopy, NMR has long been used to study the dynamics of
polymers since spin relaxation times and line-shapes are extremely sensitive to chain
motions [17-20]. Solid-state NMR experiments probe information about molecular
structure and molecular dynamics. Solid-state NMR probes chain dynamics of polymers
that are orders of magnitude slower as compared to polymers in solutions.
As mentioned above, one of the effects of radio-frequency pulses is to cause the spin
populations to deviate from their Boltzmann equilibrium population. The rate at which
Chapter 2
22
d
xation times that can be measured depending on the scale of the
relaxation time (T2) can be used to look at slower molecular
tric
the energy levels of the nuclear spin state
: quadrupolar interactions, HD: dipolar interactions, HJ :
usion [26, 27]. Both
influences on relaxation are described in the following sections.
the spin system returns to equilibrium tells us about the molecular dynamics and the
inter-nuclear distances. Molecular motions induce nuclear magnetic relaxation since
atomic fluctuations modify the local magnetic field experienced by nuclei in their
environments. These local fluctuations can occur in a broad range of frequencies an
depend directly on the chemical structure and the morphology of the material [21,22].
There are different rela
investigated motions.
Fast molecular motions can be investigated by measuring the so called spin-lattice
relaxation time (T1), or spin-lattice relaxation in the rotating frame (T1) while
measuring the spin-spin
dynamics [23, 24, 25].
For relaxation studies in more solution-like materials (e.g. amorphous polymeric system
above Tg), the relaxation times are very sensitive to molecular motions in the MHz scale
and relaxation is caused by the combination of rapid motions. In the solid-state,
relaxation times are sensitive to the same time scale of motion but also to dynamics on
the kHz scale. Depending on the type of nuclei and their environment, Dipole-Dipole
interactions (D-D), Chemical Shift Anisotropy (CSA) and QUADrupolar interactions
(for spins ≠ 1/2)(QUAD) will be responsible for the relaxation mechanism. In general,
the observed relaxation rate is the sum of all these contributions. The strength of each
interaction is very different, therefore the relaxation mechanism is mostly caused by the
interactions that are the most efficient. In quantum mechanics we will use the
Hamiltonian operator H that contains several terms describing the magnetic and elec
interactions. They lead to the splitting of s:
H = Hz +Hq +Hrf +HD +HJ (2.18)
Hz: Zeeman interactions, Hq
scalar coupling interactions.
In polymer systems, dipolar interactions (DD) between 1H-1H or 13C - 1H are the most
efficient mechanism with a strength around 100 kHz or less. Due to the weak strength,
chemical shift anisotropy of proton nuclei and the J-coupling (100 Hz) can be neglected.
Through these magnetic dipolar interactions, two majors phenomenon will dominate the
spin relaxation in polymer, molecular motions and spin diff
Chapter 2
23
2.2.2.1. Relaxation times as function of the magnetic field strength and the spectral
density of molecular motions
In NMR, transverse relaxation experiments refer to attenuation of single-quantum
coherences, which are entirely absent when thermal equilibrium is reached. As
mentioned above, the losses of coherences in polymers are mainly due to dipolar
coupling. As a consequence of molecular dynamics, the local field produced by the
dipolar interaction changes, leading to a change in the precession frequencies.
For a two-spin system the local field can be totally or partially averaged out, depending
on the frequency of the molecular motion and on the time scale on which the
information is requested,
Bloc = )1cos3(1
. ,2
,3
SISIr
(2.19)
r I,S is the inter-nuclear distance, θI,S the angle between the inter-nuclear vector and the
main field B0. Molecular motions will cause fluctuations of θI,S and depending on the
rate of the fluctuation the relaxation mechanism will be rather fast or rather slow. In
fact, the relaxation times are strongly dependent on the way the dipole-dipole
interactions are varied with molecular motions by the so called correlation time c . This
c corresponds to the time needed to describe a radius (circle divided by 2) and is
representative for the interval between successive reorientations (by vibrational,
rotational or translation motions). It is important to mention that the temperature,
viscosity and molecular weight have an influence on the correlation time c and as a
consequence also on the relaxation times [28].
Figure 2.9 shows the relationship between the relaxation times T1, T1, T2 and the
correlation time c at different field strength. One can observe that when B0 increases,
the T1 minimum increases and moves to shorter values of correlation times.
All the different orientations of the molecule will be spread over a broad range of
frequencies. The distribution of the different reorientations is given by the spectral
density function defined as:
221
2)(
c
cJ
(2.20)
Chapter 2
24
Figure 2.9. Relaxation times (T1 , T2, T 1 ) in function of correlation time c at a two magnetic field strength ( 400MHz and ---20MHz).
Bloembergen, Purcell and Pound [28] describe the dipolar relaxation rates for two equal
spins at a distance r and with isotropic fluctuations as function of the spectral density
function:
LL JJrT
24
10
316
24
1
(2.21)
LL JJJrT
22503
20
316
24
2
(2.22)
16
24
1 2
32
2
5
10
31
JJJrT LL
(2.23)
From these equations, one could see that molecular motions at the Larmor
frequency LJ (with L=0) or at twice the Larmor frequency, LJ 2 in the MHz
region, have an influence on the three relaxation times. The static term J(0) only
contributes to spin-spin relaxation and describes the interactions that cause the loss of
coherence between the spin. The 1J term is only contained in the T1 and depends
on the B1 field (kHz regime).
Chapter 2
25
The line-width of the frequency distribution in the absence of molecular motion (Δ)rl
and the correlation time c of the fluctuating interactions are used to classify molecular
motion. From equation 2.21 to 2.23 and from Figure 2.10 we may distinguish three
motional limits:
- In the complete motional averaging limit ((Δ)rl . c 1) the transverse relaxation
is influenced by non secular (rapidly varying) spin interactions when
0 . c 1. The J (0) values are low for all frequencies of molecular motions.
Therefore, none of the relaxation processes are efficient. This region is known as the
extreme narrowing limit and is located in Figure 2.9 on the left side of the minimum.
All the relaxation times are similar in that region and decrease as c increases.
The fluctuation rate of the spin interaction is fast compared to the transverse relaxation
rate and the Bloch/Wangness/Redfield (BWR) theory is used to describe the situation
(the relaxation functions are described by simple exponentials) [15]. This is generally
observed in mobile amorphous polymeric region above the glass transition, comparable
to the liquid state.
- In a rigid lattice limit and large spin system ((Δ)rl . c 1 ), the transverse
relaxation is determined by secular (low varying) spin interaction, when 0 . c 1,
which tends to be quasi-stationary on the time scale of the transverse relaxation. In this
limit, a high density of low frequency motions is observed. The spectral density at the
Larmor frequency 0 and the T1 decay, which is influenced by motions close to the
Larmor frequency, are less efficient. In this case, the J (0) and J (20) terms (see BPP
equations) are also negligible in T2 relaxation and the term J (0), corresponding to the
static term, becomes predominant, leading the spin-spin relaxation to be dominated by
the very low frequency motions, so by static dipole-dipole interactions. One of the
theories to describe this T2 behavior applied to polymer systems is the Anderson/Weiss
formalism (the relaxation functions are described by Gaussian functions) [28, 29]. This
is often observed below the glass transition in semi-crystalline polymers and in the rigid
crystalline phase.
- ((Δ)rl . c =1 ) defines partial motional averaging limit. In this limit, the J(0) is
high at the Larmor frequency, so the spin lattice relaxation is the most efficient (as seen
in Figure 2.10). The transverse relaxation is determined by both secular and non-secular
Chapter 2
26
motions. The fluctuations are fast but not fast enough to completely average out the
dipolar interactions. The proper theory to describe this situation is also found in the
Anderson/Weiss formalism, which consists of a combination of Gaussian (secular
interactions) and exponential decay functions (non secular interactions).
This limit is often observed in polymeric networks. In general, “the relaxation picture”
in polymer networks (above Tg) exhibits different molecular mobilities in the material,
producing different relaxation signals of different forms and lengths. Physical or
chemical cross-links will show rather slow molecular dynamics while dangling end or
side-chain groups will depict faster molecular dynamics. This diversity of motions
(combination of isotropic and anisotropic motions) leads to describe the relaxation with
a superposition of exponential decays or stretch exponential decay (Weybull)[30] and
Gaussian decays. However, there is still some discussion in the literature on the
mathematical form of this decay [28-32].
Figure 2.10. Plot of the spectral density J() as function of the frequency , for (a) complete motional averaging limit (b) partial motional averaging limit (c) rigid lattice limit.
A more general model that collects for both exponential and Gaussian character is
widely used to describe the relaxation of the transverse magnetization in polymer
networks:
CBcccA TtCTtBttqMTtAtM 222
22 expexp1expexp)( (2.24)
M(t) represents the magnetization function, the factor A represents the fraction of inter-
cross-link chains, B corresponds to the fraction of more mobile components such
Chapter 2
27
dangling chains, and C corresponds to the liquid-like fraction commonly named sol-
fraction. t is the pulse spacing in the pulse experiment. T2A, T2B and T2C are the related
relaxation times, c the apparent correlation time of inter-cross-link orientations, and q
a measure of the mean residual part of the second moment M2 of the dipolar interaction
in the rigid limit (T< Tg).
2.2.2.2. Spin diffusion
In polymers the 1H-nuclei are dipole coupled and will form a sort of dipole-coupled
network. If the perturbation of the magnetization occurs in one volume element, it will
diffuse to the surrounding. That is the spin diffusion phenomenon [26, 27].
Spin diffusion is defined as the transport of spin energy within the lattice by mutual
energy conservative spin flips. As in the major part of diffusion processes, this follows
Fick’s law that predicts how diffusion causes the magnetization field to change with
time; that implies the presence of a magnetization gradient.
2
2 ),(),(
r
tqMD
t
tqM
(2.25)
),( tqM describes the magnetization at the position q and at time t, D is the spin-
diffusion coefficient which depends on the average proton to proton distance as well as
on the dipolar interaction and is about 1 nm2/s in polymers [33] .
For non-isolated nuclei in heterogeneous solids, T1 and T1 relaxation times of 1H-atoms
are in different surroundings and are generally averaged out through spin diffusion. The
spin diffusion process is only efficient in a rigid lattice. In liquid-like samples, spin
diffusion is vanished because of fast isotropization of the molecular motions. In semi-
crystalline polymers, crystalline phases show a different relaxation compared to
amorphous ones, so spin diffusion will be different in the different phases. For such
reason, proton spin-lattice relaxation measurements in the rotating and laboratory frame
(1H T1 and 1H T1) are often used to estimate the lower and upper limit of domains sizes
on the basis of the following equation, valid for supposed 3D nano-domain structures
[33]:
Chapter 2
28
DTL 6 (2.26)
If the average domain size in a polymer blend is smaller than ca. 1 nm, proton-proton
spin diffusion averages out any T1 or T1 relaxation difference. In that case, all protons
decay with the same effective T1 and T1. In contrast, if the domain size is larger than ca.
50 nm, spin diffusion is too slow to average out such differences and the phases will
decay each with their intrinsic (probably different) T1 and T1 values. In the intermediate
range, 1 nm < domain size < 50 nm, we expect to find different effective T1 values and
a single effective T1. The reason is that T1 tends to be 10-100 times longer than T1, so
that spin diffusion, though unable to homogenize T1, is still able to average out T1
differences.
2.2.2.3. Summary of spin relaxation time and application in polymers
1H T1 relaxation is dominated by molecular motions at the Larmor time scale 0.1 – 10 s.
In the liquid state, spin-lattice relaxation T1 is dominated by molecular dynamic while in
solids, because of strong dipolar coupling between protons, spin diffusion is very
effective and will also determine the T1 decay. In polymers below the glass transition
(Tg), T1 will be dominated by spin diffusion (right side of the curve see Figure 2.9) and
all the 1H-nuclei will relax through spin diffusion and will depict the same T1 value,
even though we may depict different regions or phases within the polymer. Above the
Tg. molecular dynamic will also determine the relaxation (left side of the curve Figure
2.9). Generally in pure liquid, 1H T1 will be different for chemically distinct protons as
spin diffusion is vanished. Exactly the same goes for 1H T1 relaxation. However 1H T1
is affected by motions in ms scale (1 – 100 ms), so T1 is mostly shorter then T1 and
relaxation will be averaged out over smaller regions. In polymers 1H T2 relaxation is
easier to interpret, as the spin-spin relaxation time is only determined by local molecular
dynamics. 1H T2 is short 10 μs – 1 s (tenths of microseconds for polymers); therefore
spin diffusion does not have time to average out the relaxation.
Spin-lattice relaxation is mainly used for morphology study of polymeric systems; it
helps to probe information about homogeneity of mixing in a blend or in a copolymer
while spin-spin relaxation is commonly used to characterize polymeric networks (cross-
Chapter 2
29
link density, network heterogeneity) and to quantify the crystalline phase of a semi-
crystalline polymer.
Chapter 2
30
2.3. References
[1] Packer, K. J; Pope, J. M.; Yeung, R. R.; Cudby, M. E. A. J. Polym. Sci. 1984, 22, 589.
[2] Havens, J. R.; Vanderhart, D. L. Macromolecules 1985, 18, 1663.
[3] Hentschel, D. R.; Sillescu, H.; Spiess, H. W. Macromolecules 1981, 14, 1605.
[4] Schmidt-Rohr, K.; Spiess, H. W. Multidimensional Solid-State NMR and polymers, Academic Press 1994.
[5] Miller, J. B. J. Thermal Analysis 1997, 49, 521.
[6] Yu, H.; Natansohn, A.; Singh, M. A.; Plivelic, T. Macromolecules 1999, 32, 7562.
[7] Hedesiu, C.; Demco, D. E.; Kleppinger, R.; Buda, A. A.; Blumich, B.; Remerie, K.; Litvinov, V. M. Polymer 2007, 48, 763.
[8] Yao, Y.-F.; Graf, R.; Spiess, H. W. Macromolecules 2008, 41, 2514.
[9] Wind, M.; Brombacher, L.; Heuer, A.; Graf, R.; Spiess, H. W. Solid State NMR 2005, 27, 132.
[10] Demco, D. E.; Litvinov V. M.; Rata, G.; Popescu, C.; Phan, K-H.; Schmidt, A.; Blumich, B. Macromol. Chem. Phys. 2007, 208, 2085.
[11] Hedesiu, C.; Demco, D. E.; Kleppinger, R.; Poel, G. V.; Remerie, K.; Litvinov, V. M.; Blumich, B.; Steenbakkers, R. Macromol. Mater. Eng. 2008, 293, 847.
[12] Bertmer, M.; Wang, M.; Kruger, M.; Blumich, B.; Litvinov V. M.; Van Es, M.; Chem. Mater. 2007, 19, 1089.
[13] Bovey, F. A.; Nuclear Magnetic Resonance Spectroscopy, Academic Press 1988.
[14] Friebolin, H.; Basic one- and Two-Dimensional NMR Spectroscopy, VCH Wiley 1991.
[15] Bloch, F.; Hansen, W. W.; Packard, M. Phys. Rev. 1946, 69, 127.
[16] Bloch, F. Phys. Rev. 1946, 70, 460.
[17] Litvinov, V. M.; De Prajna, P. Spectroscopy of Rubbery Materials, RAPRA Technology 2002.
[18] O’Donnell, J. H.; Whittaker, A. K. Polymer 1992, 33, 62.
[19] Mohanraj, S.; Ford, W. T. Macromolecules 1985, 18, 351.
[20] Winters, R.; Lugtenburg, J.; Litvinov, V. M. Polymer 2001, 42, 24.
[21] Spyros, A. Journal of Applied Polymer Science 2003, 88, 1881.
[22] Adriaensens, P.; Storme, L.; Carleer, R.; D’Haen, J.; Gelan, J. Macromolecules 2002, 35, 135.
[23] Heuert, U.; Knorgen, M.; Menge, H.; Scheler, G.; Schneider, H. Polymer Bulletin 1996, 37, 489.
[24] Kuhn, W.; Barth, P.; Denner P.; Muller, R. Solid State Nuclear Magnetic Resonance 1996, 6, 295.
Chapter 2
31
[25] Fisher, E; Grindberg, F.; Kimmich, R. J. of Chem. Phys. 1998, 109, 846.
[26] Wang, J. J. Chem. Phys. 1996, 104, 4850.
[27] McBierty, V. J.; Douglass, D. C. J. Polym. Sci. Macromol. Rev. 1981, 16, 295.
[28] Bloembergen, E. M.; Purcell, R.; Pound, V. Physical Review 1948, 73, 679.
[29] Anderson, P. W.; Weiss, P. R. Rev. Modern. Phys. 1953, 25, 269.
[30] Litvinov, V. M.; Dias, A. A. Macromolecules 1999, 32, 3624.
[31] Simon, G.; Baumann, K.; Gronski, W. Macromolecules 1992, 25, 3629.
[32] Gotlib, Y.; Lifshitz, M. I.; Shevelen, V. A.; LishanskijI, I. S.; Balanina, V. Polym. Sci. USSR 1976 (English translation).
[33] Magusin, P. C. M. M.; Mezari, B.; Van der Mee, L.; Palmans, A. R. A.; Meijer, E. W. Macromol. Symp. 2005, 230, 126.
Chapter 3
33
Chapter 3
Morphology differences in modified polyesters probed by solid-state NMR
spectroscopy
Summary
In this study, solid-state NMR spectroscopy has been used to investigate the
morphology of poly(butylene terephthalate) (PBT)-copolymers obtained by the
incorporation of 2,2-bis[4-(2-hydroxyethoxy)phenyl]propane (Dianol) via solid-state
copolymerization (SSP). A series of such modified polyesters with varying PBT /
Dianol ratios have been investigated in previous work (M. Jansen et al.). The
morphology of the resulting products was shown to consist of a two-phase model, a
crystalline and an amorphous phase, the latter being subdivided into a rigid amorphous
fraction and a mobile amorphous fraction.
Static solid-state 1H NMR T1 measurements (spin-lattice relaxation times in the rotating
frame) were mainly used to investigate the morphology of the modified PBT samples,
together with T1(1H) -filtered 1H-13C cross-polarization experiments (CP-MAS). To be
able to distinguish between the different components of the sample, multivariate
statistics has been used. Also, from T1 (1H) and T1(
1H) measurements, upper and lower
limits of domain sizes were estimated. The results have been compared to small angle
X-Ray scattering (SAXS).
Chapter 3
34
3.1. Introduction
Poly(butylene terephthalate) (PBT) is a semi-crystalline aromatic polyester with a fast
crystallization rate. It is possible to modify PBT properties by reacting it in the melt
with other polycondensates. This modification results in a random copolymer with a
decreased crystallinity [1-6]. However, it is also possible to modify PBT while keeping
its fast crystallization rate. It has been shown in a previous study that solid-state
polymerization (SSP) is an effective method to incorporate diol co-monomers into PBT
while retaining its high crystallization rate and enhancing other properties such as
mechanical properties [7].
The process of SSP is to operate at a reaction temperature just below the melting
temperature of the homopolymer. At this temperature, the polymer-chain segments in
the amorphous phase are mobile enough allowing trans-esterification to occur between
the homopolymer and the co-monomer [7, 8, 9]. However, the polymer-chain segments
in the crystalline phase are insufficiently mobile to participate to the reaction and will
remain unchanged.
This chapter looks at the morphology of modified PBT samples with a Dianol co-
monomer 2,2-bis[4-(2-hydroxyethoxy)phenyl]propane (commonly named Dianol 220)
obtained via SSP by means of solid-state NMR. A series of these modified polyesters has
been investigated with varying Dianol / PBT ratios. These samples were prepared by M.
Jansen [10] who previously showed that the morphology of the resulting products
follows a (two)-phase model, composed of a crystalline phase and an amorphous phase,
the latter subdivided into a rigid amorphous fraction and a mobile amorphous fraction.
Jansen et al. [10] developed a special calculation method showing that only the mobile
amorphous fraction is accessible for incorporation of the Dianol via SSP and that Dianol
is randomly incorporated into this mobile amorphous fraction. Their conclusions were
based on Differential Scanning Calorimetry (DSC) measurements and computation of
the degree of randomness using liquid-NMR spectroscopy.
The properties of materials are strongly related to their morphology. A variety of
techniques, such as thermal analysis, mechanical relaxation, microscopy, light scattering
and computer simulation are available to probe the heterogeneity of multiphase systems.
Each method has its own advantages and disadvantages.
Chapter 3
35
The use of solid-state NMR for investigating morphology and semi-quantitatively
estimate the microstructure of the different phases present in the system have been
demonstrated by many authors [11-22].
The results discussed below illustrate the use of solid-state NMR relaxometry for
investigating nano-structures (morphologies) and dynamics of the PBT modified
samples. The incorporation of the Dianol in the mobile amorphous fraction was also
investigated. Static solid-state NMR measurements of 1H spin relaxation (spin-lattice
relaxation time in the rotating frame, T1) were used to investigate the morphology of
the modified PBT samples, together with T1(1H) -filtered 1H-13C cross-polarization
experiments (CP-MAS). The latter makes it possible to observe the chemical shifts and
the relaxation parameters for each resolvable carbon nucleus and their protons. The
observed chemical shift provides information concerning the conformation and packing
of the polymer while the relaxation parameters indicate its dynamics. Also, upper and
lower limits of domain sizes were established. The results have been compared to small-
angle X-Ray scattering (SAXS).
T1 relaxation measurements probe molecular motions characterized by frequencies in
the mid-kilohertz region. These measurements are used to study phase separation in
semi-crystalline polymers as they allow more spatial averaging then T2 relaxation times.
The molecular motions in a crystalline phase are much more restricted then in an
amorphous matrix. At a temperature between the glass transition temperature (Tg) and
the melting point (Tm) there is a big contrast between the mobility in the crystalline
phase and in the amorphous phase. These two types of molecular structures will
therefore demonstrate different molecular dynamics, hence different T1 relaxation
times [23-28]. Such measurements probe quantitative information, as in general the area
underneath the solid-state NMR signal is proportional to the number of protons in the
system. In that way, it is possible to quantify the number of proton in each phase.
3.2. Experimental Section
3.2.1. Sample preparation
The modified PBT samples were obtained and synthesized by M.A.G. Jansen [10].
Chapter 3
36
The synthesis of PBT-Dianol copolymers by SSP is described in Jansen’s thesis.
Different ratios of PBT and Dianol were used and are reported in Table 3.1. BDx
corresponds to the fraction in mol% of PBT (expressed in 1,4-butanediol units) whereas
Diy corresponds to the fraction in mol% of Dianol in the initial mixture used for the SSP
(BDxDiy)feed . After SSP, the mol fractions slightly differ. This difference may be
attributed to the evaporation of 1,4-butanediol or ethylene glycol during the SSP
reaction. The general chemical structure of the (BDxDiy)ssp copolymer samples is shown
in Figure 3.1:
Diol
T = 180-200 °C
N2
Crystalline part
Amorphous part
DiolDiol
T = 180-200 °C
N2
T = 180-200 °C
N2
Crystalline part
Amorphous part
HOCH2CH2O C
CH3
CH3
OCH2CH2OHCC
O O
OCH2(CH2)2CH2O
n
C
CH3
OCH2(CH2)2CH2O-C OCH2CH2O
o o
-C-
CH3
-OCH2CH2O-C C
o o
m n
C
CH3
OCH2(CH2)2CH2O-C OCH2CH2O
o o
-C-
CH3
-OCH2CH2O-C C
o o
m n
C
CH3
OCH2(CH2)2CH2O-C OCH2CH2O
o o
-C-
CH3
-OCH2CH2O-CC
CH3
OCH2(CH2)2CH2O-C OCH2CH2O
o o
-C-
CH3
-OCH2CH2O-COCH2(CH2)2CH2O-C OCH2CH2O
o o
-C-
CH3
-OCH2CH2O-C C
o o
m n
C
o o
m n
(a) (b)
(c)
(d) (e)
Diol
T = 180-200 °C
N2
Crystalline part
Amorphous part
DiolDiol
T = 180-200 °C
N2
T = 180-200 °C
N2
Crystalline part
Amorphous part
HOCH2CH2O C
CH3
CH3
OCH2CH2OHCC
O O
OCH2(CH2)2CH2O
n
C
CH3
OCH2(CH2)2CH2O-C OCH2CH2O
o o
-C-
CH3
-OCH2CH2O-C C
o o
m n
C
CH3
OCH2(CH2)2CH2O-C OCH2CH2O
o o
-C-
CH3
-OCH2CH2O-C C
o o
m n
C
CH3
OCH2(CH2)2CH2O-C OCH2CH2O
o o
-C-
CH3
-OCH2CH2O-CC
CH3
OCH2(CH2)2CH2O-C OCH2CH2O
o o
-C-
CH3
-OCH2CH2O-COCH2(CH2)2CH2O-C OCH2CH2O
o o
-C-
CH3
-OCH2CH2O-C C
o o
m n
C
o o
m n
Diol
T = 180-200 °C
N2
Crystalline part
Amorphous part
DiolDiol
T = 180-200 °C
N2
T = 180-200 °C
N2
Crystalline part
Amorphous part
HOCH2CH2O C
CH3
CH3
OCH2CH2OHCC
O O
OCH2(CH2)2CH2O
n
C
CH3
OCH2(CH2)2CH2O-C OCH2CH2O
o o
-C-
CH3
-OCH2CH2O-C C
o o
m n
C
CH3
OCH2(CH2)2CH2O-C OCH2CH2O
o o
-C-
CH3
-OCH2CH2O-C C
o o
m n
C
CH3
OCH2(CH2)2CH2O-C OCH2CH2O
o o
-C-
CH3
-OCH2CH2O-CC
CH3
OCH2(CH2)2CH2O-C OCH2CH2O
o o
-C-
CH3
-OCH2CH2O-COCH2(CH2)2CH2O-C OCH2CH2O
o o
-C-
CH3
-OCH2CH2O-C C
o o
m n
C
o o
m n
(a) (b)
(c)
(d) (e)
Figure 3.1. (a) Dianol monomer (b) poly-(butylene terephthalate) (PBT) (c) poly-(butylene terephthalate) (PBT)-copolymers chemical structures; incorporation of diols in PBT via solid-state polymerization, before SSP(d) a physical mixture of PBT and dianol; after SSP(e), PBT copolymer with crystallizable PBT blocks [7].
Chapter 3
37
The original PBT pellets (Mn=15 kg/mol Mw = 34 kg/mol) were provided by GE
Plastics (The Netherlands) and the 2,2-bis [4-(2-hydroxyethoxy)phenyl]propane (Dianol
220) was provided by Air Liquide (France). PBT and Dianol were mixed by dissolution
in 1,1,1,3,3,3-hexafluoro-2-propanol (HFIP) prior to SSP. For all SSP reactions, a
reaction time of 9 hours was used at a reaction temperature of 180 C. Average
molecular weights Mn were obtained using size exclusion chromatography (SEC) [1].
3.2.2. NMR experiments
1H-decoupled 13C NMR spectra were recorded on a Bruker DMX500 spectrometer
using a CP/MAS probe with the observe channel tuned on 13C and the decoupler
channel to 1H. All 13C NMR spectra were recorded by use of standard 1H-13C cross-
polarization with an mplitude modulated contact pulse of 1ms and 5 s delay time.
T1(1H)- filtered CP-MAS spectra were measured with a /2 proton lock pulse
preceding the 1H-13C contact pulse; magic angle spinning was performed at 12 kHz.
Static T1 decays were measured with a /2-spin-lock-hahn-echo sequence at 353 K and
435 K. Some further experiments on samples BD96Di04 and BD54Di46 have been carried
out at U-Hasselt, at ambient temperature, 353 K and at 435 K on a Varian Inova 400
spectrometer using a dedicated wide-line probe equipped with a 5 mm coil. For T1 all
experimental results could be fitted using a mono-exponential decay. For T1
experiments, two and three exponential decays were used. However, the fit with three
components was found to be significantly better.
3.2.3. NMR Pulse sequence
3.2.3.1. Inversion recovery
To determine the proton spin lattice relaxation time T1, the inversion recovery pulse
sequence has been used. The equilibrium magnetization is first disturbed by a 180z rf-
pulse. As a consequence, the magnetization is inverted along the longitudinal direction
(Figure 2.6.b in Chapter 2). Due to the T1 relaxation, the magnetization becomes
smaller, passes through zero and starts to grow back to its original value. However,
Chapter 3
38
igure 3.2. T1(1H)- filtered CP-MAS-pulse sequence
longitudinal magnetization cannot be detected directly, therefore a 90 pulse is applied
for detection.
3.2.3.2. /2-spin-lock- pulse sequence
To determine the static proton spin lattice relaxation time in the rotating frame T1, a
classical 90x’ pulse has been used followed by a 90 phase shift of the transmitter (see
Chapter 2, Figure 2.8). In such a way, the magnetization is spin-locked along the y’-
axis. The B1 field strength used in our experiments is 50 kHz.
3.2.3.3. T1(1H)- filtered CP-MAS-pulse sequence
The pulse sequence starts also with a protons /2-spin-lock pulse sequence. The
magnetization of the proton is spin-locked during a time t. After the time t, the carbons
are brought into contact with protons; this is the cross-polarization. During that time, the
proton magnetization is transferred to the carbon spin system. Afterwards, the carbon
signal is detected under high-power proton coupling transition (Figure 3.2).
F
Chapter 3
39
3.2.4. NMR Fitting procedure (2D analysis)
The NMR signal is acquired as a function of time t (spin locking time):
(3.1)
where Ginhom(
)exp()2|()(),2( 000inhom0 titRGdtS 0 ) reflects the inhomogeneous spectrum arising from residual
anisotropy, chemical shift differences and magnetic field inhomogeneity; R( 0 |2 + t)
is the spin relaxation decay associated with spectral frequency ω0. In the usual approach,
the maximum of the signal at t = 0 is measured as a function of 2 without spectral
resolution. The resulting monotonous decay, a 1D data set, can be analyzed in terms of a
number of exponential components. It is proposed to use the spectral resolution in the
full 2D data set I (2, ) after Fourier transformation in the t dimension (see detail of
2D analysis in Chapter 5). For a mixture of components with mono-exponential
relaxation and well-separated spectral frequencies and relaxation rates, it should be
relatively easy to extract the intrinsic properties from a series of the spectra. In real
cases, however, components may overlap in both the spectral and Laplace domain. To
be able to distinguish between the different components of the sample, a data analysis
approach based on multivariate statistics has been used. This approach uses a
“Multivariate Analysis” to estimate the line shapes of the overlapping proton resonance.
There are several methods to fit the exponential decay model, the one used here, is
based on the so-called LINOL method describes in Appendix 2 (Linear-Non-Linear
fitting).
3.2.5. Software
Home-built routines, written in MATLAB 6.5 (The Mathworks, Natick, MA, USA),
were used for NMR data treatment. The solid-state NMR relaxation data were fitted
using a combination of exponential decay functions with the use of a novel procedure
based on multi-variable analysis.
3.2.6. Modulated DSC
It was shown that modulated DSC (MDSC) allows for deconvolution of signals from
amorphous and crystalline phases. The method has a greater resolution then non-
Chapter 3
40
modulated DSC. Amorphous phases give rise to a reversing heat flow, which is readily
converted into a specific heat capacity (Cp) and its derivative, which highlights the glass
transition temperatures. The Tg’s and the corresponding heat capacities of the modified
PBT samples were measured via (MDSC) using TA Q1000 DSC, equipped with an
auto-sampler and a refrigerated cooling system (RCS). The temperature calibration was
obtained using Indium. The (BDxDiy)ssp copolymer samples were prepared in aluminum
pans (between 10 and 15 mg). An oscillating heat flow signal with a period of 60
seconds and amplitude of 1 C and an underlying heating rate of 1 C/ min was used.
All samples were measured following a specific ageing procedure. The samples were
first heated in normal DSC mode from 0 C to 240 C then cooled and heated up again
in modulated DSC to 140 C. Each sample was kept at this temperature during
respectively 1 hour, 10 hours and 20 hours. After each ageing time, the samples were
cooled to 0 C and reheated using a cooling and heating rates of 10 C/min. The Tg
values were extracted from the signal of the reversing heat flow.
3.3. Results and discussion
3.3.1. 13C-CP-MAS experiments
Figure 3.3 shows 13C-CP-MAS spectra of the PBT homopolymer, Dianol monomer and
PBT copolymer. By comparison it is easy to assign each peak from the copolymer
sample. Generally, the narrow shapes correspond to the PBT while the broader shapes to
the Dianol.
In Figures 3.4 and 3.5, 13C–CP-MAS spectrum recorded at 353 K temperature at
different spin-locking times are shown. The 13C detection allows signals of PBT and
Dianol to be resolved on the basis of chemical shift whereas the decay of the signal
intensity as function of the spin-locking time reflects T1H of each component. At longer
spin-locking time (filter time) the signals corresponding to the most mobile phase of the
polymeric system are removed while the ones corresponding to the more rigid phase of
the system remain. T1(1H)- filtered CP -MAS experiments were recorded after the
application of a /2 pulse on the 1H channel, followed by a variable spin-locking time
then by 13C detection through 13C 1H CP at fixed contact time at 353 K.
Chapter 3
41
200 150 100 50 0 ppm
(a)
(b)
(c)
200 150 100 50 0 ppm
(a)
(b)
(c)
Figure 3.3. 1H- 13C CPMAS spectra of the PBT homopolymer (a), Dianol monomer (b), PBT copolymer (BD54Di46)ssp (c), measured at room temperature .
From the T1(1H)-filtered 13C-CP-MAS spectra of BD76Di24 in Figure 3.4 one can see
that the first spectrum recorded at a filtered time of 10 s contains the signals of both
PBT and Dianol. The main rigid PBT-component stays equally divided over the
complete spectrum while the resonance peaks of the Dianol are decreasing with
increasing filter time. At longer filtered times one can also notice a line-narrowing of
the overlapping peak at 125 ppm.
In Figure 3.5, the spectrum of BD54Di46 at filter time 10 s clearly shows separate
Dianol peaks at about 30 and 115 ppm. Increasing the filter time to 6 ms shows a
disappearance of the Dianol peaks. Also the increase in signal to noise is a consequence
of the increased molecular mobility of the BD54Di46 copolymer.
Chapter 3
42
200 150 100 50 0 ppm
(a)
(b)
(c)
200 150 100 50 0 ppm200 150 100 50 0 ppm
(a)
(b)
(c)
Figure 3.4. T1 -filtered 1H- 13C CPMAS spectra of the copolymer (BD76Di24)ssp at filter times 10 s (a), 3ms (b), 6 ms (c), measured at 353 K.
T1(1H)-filtered 13C-CP-MAS experiments show two types of chains mobility in the
material. The Dianol signals show a relaxation time T1 of about 2 ms. Cross-
polarization tends to overemphasize the rigid polymer phase in 13C - NMR spectra, so
only 1H with sufficiently long T1 values could be detected given the presence of a finite
contact time within the pulse. Since the filtered spectra are based on 1H -13C cross-
polarization, a quantitative analysis is not possible. For more quantitative analysis, static
proton spin-lattice relaxation in the rotating frame T1 experiments were carried out.
The measurements of 1H spin-lattice relaxation in the rotating frame are very sensitive
to motions in the mid-kilohertz region and typically involve main chain segmental
motions of polymers above the glass-transition temperature. Also, such measurements in
principle probe quantitative information, as the area underneath the solid-state NMR
Chapter 3
43
-50250 200 150 100 50 0 ppm
-50250 200 150 100 50 0 ppm
a
b
-50250 200 150 100 50 0 ppm
-50250 200 150 100 50 0 ppm
-50250 200 150 100 50 0 ppm
-50250 200 150 100 50 0 ppm
a
b
(a)
(b)
-50250 200 150 100 50 0 ppm
-50250 200 150 100 50 0 ppm
a
b
-50250 200 150 100 50 0 ppm
-50250 200 150 100 50 0 ppm
-50250 200 150 100 50 0 ppm
-50250 200 150 100 50 0 ppm
a
b
(a)
(b)
Figure 3.5. T1 -filtered 1H- 13C CPMAS spectra of the copolymer (BD54Di46)ssp at filter times 10 s (a) and 6 ms (b), measured at 353 K.
line is proportional to the amount of protons in the system. In that way, it is possible to
quantify the number of protons in each phase of the system.
3.3.2. 1H T1 and 1H T1 relaxation experiments
NMR spin-lattice relaxation times are not only determined by molecular dynamics but
also by spin diffusion where energy is transferred throughout the spin system by
successive spin flips. The NMR spin-lattice relaxation time measurements T1 and T1
are generally used to identify heterogeneity of materials at different levels, T1 at ~ 5 to
>100 nm size and T1 at 1 to 5 nm size. They provide information about the miscibility
and the domain size of various phases inside the polymer system [29-32].
cs but
also by spin diffusion where energy is transferred throughout the spin system by
successive spin flips. The NMR spin-lattice relaxation time measurements T1 and T1
are generally used to identify heterogeneity of materials at different levels, T1 at ~ 5 to
>100 nm size and T1 at 1 to 5 nm size. They provide information about the miscibility
and the domain size of various phases inside the polymer system [29-32].
T1 experiments were first used in this work to semi-quantitatively determine the micro-
domain structures of the materials, based on mobility differences between the crystalline
phase, the rigid amorphous phase and the mobile amorphous phase. T1 values were
T1 experiments were first used in this work to semi-quantitatively determine the micro-
domain structures of the materials, based on mobility differences between the crystalline
phase, the rigid amorphous phase and the mobile amorphous phase. T1 values were
Chapter 3
44
obtained from the analysis of the decay of the magnetization measured at increasing
spin-lock times after the initial spin-locking excitation pulse. According to the
Bloembergen, Purcell and Pound theory (BPP), rigid phases in polymers should have
protons depicting a longer T1 relaxation time than mobile ones [33]. This has been
verified at room temperature for sample BD96Di04 but was not observed in the sample
containing the highest mole fraction of Dianol (BD54Di46)ssp. This is due to a very high
mobility in the sample.
To be able to distinguish between the two amorphous fractions, we recorded data at 353
K and 435 K, (30 and 110 oC) above the Tg of the samples. For the T1H experiments, the
values of the relaxation times decrease at increasing temperatures (at RT and at 162 °C
from 1.15 s to 0.86 ms). For all the samples, only one component fit was needed to fit
the data. The T1H values are located at the right side of the relax–curve of logT1 as a
function of log c (see also Figure 2.9). The relaxation decay is dominated by spin
diffusion longer then the domain sizes.
Log c
Log T 1
Log c
Log T 1
Figure 3.6. Spin-lattice relaxation time in function of correlation time, and distribution of different (BDxDiy)ssp fractions (rigid-solid phase (), intermediate (), mobile (). Plain markers correspond to high amount and hollow ones to low amount).
Chapter 3
45
For all T1 experiments, the data were fitted best with a three components fit at both
temperatures and for all the samples containing Dianol. The T1 values are found to be
much longer for protons in the mobile phases than for the ones in the rigid phase. This
has been verified by recording the data at low and high temperatures. The T1 values for
the samples increase when temperature increases, suggesting at first that the values are
located on the left side of the curve in Figure 3.6. For the proper interpretations of the
T1 values, however, one should mention that during the relaxation in ms, spin diffusion
takes place over a limited distance (few nm). Effectively, spin diffusion will take place
with protons belonging to the same environment (same unit or monomer) and the same
phase, but will only partly reach the protons in the other phases. Even in the rigid
phases, spin diffusion as well as local molecular mobility influence the relaxation decay.
In the most mobile part, mainly local molecular mobility influences the decay. In each
phase, there are mobile protons (xf) and less mobile protons (xs). The T1 values are
dominated by the fastest relaxing protons according to the equation (3.2):
ρT1
1 =
ρs
s
ρf
f
T
x
T
x
11
(3.2)
The proposed T1 values are all at the left side of the relax minimum. In Figure 3.6 we
propose an interpretation for the T1 positions that will also depend on the amount of
fast and slow protons in each fraction.
At 435 K, the best fit of the decay is found to be multi-exponential. One and two phases
decay models perform significantly worse. A three-component model gives a good fit as
shown in Figure 3.7.
Three different values of T1 could be extracted. These three values correspond to three
types of 1H mobility present in the system. The obtained T1 values differ by one to two
orders of magnitude as shown in Table 3.2, the times 0.4, 0.05 and 34 ms would
correspond to an intermediate rigid phase (rigid amorphous phase), the values 10, 1.2, 3
and 0.9 ms to the most immobile solid phase and the longest T1 decays from 12 to 161
ms corresponding to a most mobile amorphous phase. The overall mobility is in good
agreement with Tg values from MDSC results presented in Table 3.1.
Chapter 3
46
Figure. 3.7. T1 relaxation decay curves recorded at 435K (experimental data fitted with three exponential decay model). () correspond to sample (BD100Di00)ssp; () correspond to sample (BD54Di46)ssp ; () correspond to sample (BD96Di04)ssp
Table 3.1. Glass transition temperature (Tg) for the synthesized (BDxDiy)ssp copolymers obtained from MDSC measurements.
(BDxDiy)feed mixtures
(BDxDiy)ssp copolymers
Heating run 1
x y FBD-T, total FDi-T, total Tg
[mol%] [mol%] [mol%] [mol%] [°C]
100 0 100 0 53.5
95 5 96 4 62
80 20 76 24 67.8
65 35 54 46 51
Chapter 3
47
Table 3.2. T1 relaxation times obtained with 3 exponential decay fit at 435 K using 2D data analysis.
FBD-T, total
[mol%] FDi-T, total
[mol%] Crystalline T1i (ms)
Rigid amorphous T1s (ms)
Mobile amorphousT1l (ms)
100 0 10 0.4 45
96 4 1.2 0.05 12
76 24 3 0.05 52
54 46 0.9 34 161
FBD-T, total
[mol%] FDi-T, total
[mol%] Fraction
(%) Fraction
(%) Fraction
(%)
100 0 23 51 26
96 4 7 26 67
76 24 3 35 62
54 46 19 39 42
For all modified PBT samples, three ranges of mobility are observed with similar
relaxation times, except for the sample with the highest concentration of Dianol
(BD54Di46)ssp. This sample shows much higher relaxation time values for the
intermediate and mobile phase while the solid phase has a normal low T1 value of 0.9
ms, close to the most solid phase T1 values of the other copolymers. These T1 values
can be explained by the higher Dianol incorporation in the original PBT non-crystalline
fraction. Also, one can see from Figure 3.8 of the well-resolved 1H NMR spectrum of
the most mobile phase that the proton peaks show high T1 values. There is still
sufficient peak intensity at the long proton locking time of 240 ms.
Table 3.3 shows two components analysis of the proton spectra of the sample
(BD54Di46)ssp at 435 K, resulting in well-separated short and long proton T1 values for
each of the four proton signals. The aromatic peak around 7 ppm and the aliphatic peak
around 4 ppm are representative for the PBT as well as for the Dianol regions while the
signal at 1.6 ppm mostly shows the Dianol domains, represented by the Me groups (part
of it is also due to the CH2 group). A two components fit of the signal at 1.6 ppm results
in T1 values of 24 and 145 ms, corresponding respectively to an intermediate solid
amorphous and a mobile amorphous phase. These values closely resemble the values of
the three components fit in Table 3.2 of 34 and 161 ms respectively for the intermediate
and the mobile phase. The two component analysis of the other three proton signals,
Chapter 3
48
representative for PBT as well as for the Dianol, show a low T1 value around 1 ms for a
solid phase and a high T1 value between 58 and 107 ms for a mobile phase. From the
signal at 1.6 ppm, a value much higher than 1ms is found for T1. This confirms that
during SSP the Dianol monomer is only incorporated in the amorphous phase of the
PBT.
Figure 3.8. NMR signal of the (BD54Di46)ssp copolymer recorded at 435 K using T1 s
relaxation experiments obtained at t = 240 ms (thin line) t = 10 s (broad line).
Table 3.3. T1 relaxation times obtained via peak peaking analysis with 2 exponential decay fit for the sample (BD54Di46)ssp containing the highest amount of Dianol at 435 K.
ppm T1 s (ms) T1l (ms)
7.8 1 87
6.9 0.65 107
4.3 1 58
1.6 24 145
Chapter 3
49
All the NMR relaxometry data have been analyzed with respectively 1D data fits and
2D data fits. For all samples, the 2D analyses seem to show consistency with reality and
are close to the value found by M. Jansen [10]. For the sample (BD54Di46)ssp, the 1D
analysis was preferred. Effectively, the mobility in the sample is much higher and the
resolution drastically increases (already observed in Figure 3.8). The fact that the 2D
analysis takes the chemical shift into account may not always reflect the phase
relaxation. It may also assimilate relaxations of protons in a phase such as side groups
(Me-) to the overall protons relaxation in that same phase relaxation. The 1D data seem
to be closer to the physical reality and also close to what has been obtained using a
dedicated wideline probe with the 400 MHz NMR spectrometer.
Also, one additional argument, showing the strong influence of the Me-groups
contained in the Dianol, is that between 1D and 2D analysis the data are very close for
the most rigid phase while for the intermediate and mobile one, the relaxation times are
different (Table 3.4).
Sample (BD54Di46)ssp contained much more Dianol, so one expects a higher chain
mobility as clearly observed from the much lower Tg value, resulting from MDSC
measurements (Table 3.1), as compared to the other modified samples. This has already
been reported by M. Jansen et al. and is due to the lower number-average molecular
weight (Mn = 8 kg/mol) of the polymer chains [7].
Table 3.4. Rigid fraction, intermediate fraction (rigid amorphous) and mobile fraction obtained from static T1 experiments with 3 exponential decay fit at 500MHz, 400 MHz(1D analysis and 2D analysis).
FBD-T, total
[mol%]
FDi-T, total
[mol%]
Crystalline
T1s (ms)
Rigid
amorphous
T1i (ms)
Mobile amorphous
T1l (ms)
2D (500MHz) 54 46 0.9 34 161
1D (500MHz) 54 46 1.1 23.1 77
1D (400MHz) 54 46 2 9 28
FBD-T, total
[mol%]
FDi-T, total
[mol%] Crystalline
(%)
Rigid amorphous
(%)
Mobile amorphous
(%) 2D (500MHz) 54 46 19 39 42
1D (500MHz) 54 46 9.7 31 50.4
1D (400MHz) 54 46 6.5 38.5 55
Chapter 3
50
Some experiments have also been carried out at 353 K and the obtained T1 values were
compared to the ones found with T1(1H)- filtered CP-MAS experiments. Fitting the
system with a two-exponential model gives a T1 value for the mobile component of
about 3 ms, very close to the one found for the mobile component in the CP-
experiments (2 ms). For all samples the values were in good agreement with those found
using CP. However, the best fit also here was obtained with a three-exponential model
and the T1 value for the most-mobile component was found to be about 10 ms. A
mobility higher than the one found in CP experiments for Dianol supports the
suggestion that the Dianol is not only present in the most-mobile fraction of the system
but also in the intermediate phase (CP is effective for less mobile protons).
Some further experiments have been obtained on a NMR spectrometer at 400 MHz, for
the sample with the least fraction of Dianol and the most fraction of Dianol.
In both cases the best fit was also obtained using a three- exponential decay fit. For
sample (BD96Di04)ssp the results obtained are in good agreement with what has been
obtained using the 500 MHz system. For sample (BD54Di46)ssp discrepancies are
observed, probably due to the influence of the chemical shift on the analysis method to
fit the data (Table 3.4). The number of experimental points taken for experiments will
probably also have an influence of the results.
3.3.3. Quantification of each phase
As already mentioned above, static proton T1 experiments can yield more quantitative
information, as this technique does not involve cross-polarization or magic-angle
spinning. The amount of 1H in each phase was quantified (assuming that for most
samples, the system is still in the diffusion limit with low spin diffusion for the more
mobile amorphous samples). These fractions are reported in Table 3.2 and Table 3.4,
and are obtained with a multiple-exponential decay fit using (for most of them)
multivariate analysis (Appendix 1 and 2):
lis TtCTtBTtAtM expexpexp)( (3.3)
M (t) represents the magnetization function, A represents the fraction of components
with a short relaxation time, B corresponds to the fraction of components with an
Chapter 3
51
intermediate relaxation time and C corresponds to the fraction of components with long
relaxation time; t is the spin lock time.
So far, no much correlation regarding phase quantification has been found between
these data and these obtained by Jansen et al. via DSC and solution NMR reported in
Table 3.5.
In both cases, models are used and different assumptions are made. Also different
techniques probe different information, so that quantification remains a challenge. The
crystallinity obtained by Jansen et. al. was determined via DSC and is thus based on the
amount of heat needed to break the crystal order, whereas the crystallinity determined
with NMR is based on differences in chain mobility within the rigid crystalline phase
and the amorphous one (besides NMR relaxation probes proton fractions while DSC
reports the data to mole fractions). Also T1 relaxation decay is influenced by both spin
diffusion and local molecular dynamic. This will clearly affect the physical
interpretation of each phase in the system and thereby the relative fraction of proton in
each phase. However, Table 3.2 shows that the crystalline fraction is decreasing with
increasing amount of Dianol. The Dianol acts as a reactant for the trans-esterification
reaction in the amorphous phase but also as a solvent (swelling agent) for that same
phase. Furthermore, during the time scale of NMR experiments, some rearrangement in
the amorphous phase may occur, that could explain some of the observed differences.
Table 3.5. (%) of the three fractions present in the system derived from MDSC and solution NMR (Martijn Jansen et al.).
FBD-T, total
[mol%] FBD-T, total
[mol%] Crystalline
(%) Rigid amorphous
(%) Mobile amorphous
(%)
100 0 54 28 18
96 4 50 22 28
76 24 32 8 60
54 46 8 1 91
The samples were then measured at room temperature then heated to 435 K and cooled
down to room temperature and measured again. No significant differences were
observed in the fraction of components and in the relaxation times. In addition,
Chapter 3
52
modulated MDSC experiments have been carried out and all samples were measured
following a specific ageing procedure. The variations in the heat capacity are shown in
Table 3.6. The results depict no variation in heat capacity indicating that no
rearrangement in the amorphous fraction takes place.
Some complementary analyses have been done to help estimate the amount of rigid and
mobile 1H-atoms contained in each different phases of the system. Deconvolution and
integration of the 1H-NMR signal corresponding to the broad part of the raw signal at a
short spin locking time obtained from T1 experiments were done. The results are
reported in Table 3.7 and have been compared to the total rigid fraction found by Jansen
et al (the total rigid fraction being the sum of the crystalline fraction and the rigid
amorphous fraction). The integration of the deconvoluted signal gives values for the
rigid component, very close to those found by Jansen et al. [7,10].
Table 3.6. TMDSC results obtained before and after ageing procedure.
(BDxDiy)ssp copolymers Heatingrun1 No
ageing
Heating run 2 after 1 hour
ageing
Heating run 2
after 10 hours
ageing
Heating run 2
after 20 hours ageing
Cp Cp Cp Cp FBD-T, total
[mol%]
FDi-T, total
[mol%] J/g C
100 0 0.39 0.4 0.39 0.38
96 4 0.42 0.41 0.37 0.40
76 24 0.47 0.48 0.50 0.47
54 46 0.58 0.53 0.59 0.6
If we look at Table 3.8 showing the static NMR line width analysis by peak
deconvolution and integration of the sample (BD54Di46)ssp, we can observe that the
relative area underneath the peak at 1.6 ppm is the same at a spin locking time of 10 s
and at a spin locking time of 240 ms. This clearly suggests that Dianol is not only
contained in the most mobile part. The relative area underneath the peak at 1.6 ppm,
mostly corresponding to the Me-protons of Dianol, would have a much higher value at a
Chapter 3
53
spin locking time of 240 ms, typical for the most mobile amorphous. The two phases
behavior of Dianol has also been noticed before, showing T1 values of 24 ms
(intermediate phase) and 145 ms (mobile phase) in Table 3.3.
Table 3.7. Rrigid fraction obtained by integration of the deconvoluted peak corresponding to the broad part of the raw signal.
FBD-T, total
[mol%]
FDi-T, total
[mol%] Rigid fraction
(%)
100 0 84
96 4 84
76 24 44
54 46 5
Table 3.8. Peak deconvolution and integration from the static line width analysis of sample (BD54Di46)ssp obtained at 435 K and at two different spin locking time.
ppm Area (%)
at 10 s
Area (%)
at 240 ms
7.8 17.5 18
6.9 18.3 26
4.3 31 27
1.6 28 29
3.3.4. Domain sizes
As already mentioned above, proton spin-lattice relaxations in rotating and laboratory
frame measurements (1H T1 and 1H T1) provide information about the level of
heterogeneity (phase morphology) on the nm scale due to the process of proton spin
diffusion [34]. This latter process is not a physical movement of protons but a transfer of
spin energy by successive spin-flips. This process is often modeled as Fickian diffusion.
It allows to judge the degree of size separation and to estimate the size of molecular
domains. In polymeric systems proton spin-lattice relaxations in rotating and laboratory
Chapter 3
54
frame measurements (1H T1 and 1H T1) are often used to estimate lower and upper limit
of domain sizes on the basis of the following equation, valid for lamella structures [35-
37]:
16 TDL s (3.4)
Ds is the spin-diffusion coefficient which depends on the average proton- to- proton
distance as well as on the dipolar interaction.
If the average domain size in a polymer blend is smaller than ca. 1 nm, proton-proton
spin diffusion averages out any T1 or T1 relaxation differences. In that case, all protons
decay with the same effective T1 and T1. In contrast, if the domain size is larger than ca.
50 nm, spin diffusion is too slow to average out such differences and the different
phases will decay each with their intrinsic (probably different) T1 and T1 values. In the
intermediate range, 1 nm < domain size < 50 nm, we expect to find different effective
T1 values and a single effective T1. The reason is that T1 tends to be 10-100 times
longer than T1, so that spin diffusion is thought to be unable to homogenize T1, but is
still able to average out T1 differences. If we look at the SAXS results obtained by
Jansen et al. (5 nm < domain size < 50 nm) shown in Figure 3.9 and the data obtained
through NMR relaxation experiments, we seem to be in good agreement with the above
concept.
Also, T1 values were obtained from the analysis of the recovery of the magnetization
measured at increasing times after the initial inversion pulse. For all samples and at each
studied temperature, the recovery was found to be mono-exponential indicating that spin
diffusion averages the T1 characteristic of protons belonging to different domains.
Assuming a spin-diffusion coefficient Ds of 1 nm2 /ms [26] and considering that the
minimum T1 observed for all samples being about 1 s an upper limit of about 80 nm
based of equation (3.4) is proposed. A lower boundary could also be estimated using the
values of T1 in the rigid phase. The data are reported in Table 3.9. An upper and lower
limit for the mixing scale (domain sizes) may give some idea about the size of the
crystalline and the (rigid)-amorphous domains. The limits in the rigid fractions are in
reasonable agreement with those obtained using SAXS by Jansen et al. [7].
Chapter 3
55
Figure 3.9. Morpbhological parameters obtained from SAXS: long period Lp, average crystal thickness lx, average amorphous layer thickness l ( M. Jansen et al. ).
Table 3.9. Lower boundary of domain size obtained from T1 of the three fractions obtained at 435 K (Li corresponding to the lower boundary limit of the intermediate fraction, Lr corresponding to the lower boundary limit of the rigid fraction, Lm corresponding to the lower boundary limit of the mobile fraction).
FDi-T, total
[mol%]
Li
(nm)
Lr
(nm)
Lm
(nm)
0 1.54 7.7 16.4
4 0.54 2.68 8.48
24 0.54 4.24 17.66
46 7.34 3.4 12.96
3.4. Conclusions
The Dianol was first identified in CP-MAS experiments. CP-MAS experiments are
sensitive only in a rigid limit (because of cross-polarization efficiency) so it follows that
Dianol is not only incorporated in the most-mobile fraction of the sample but also in the
amorphous interphase.
Chapter 3
56
For all samples, 1H T1 experiments do not exhibit a single or dual-exponential decay
behavior, but a much better fit was found using a three exponential decay fit. Such a
result is in good agreement with the three fractions (two phases) demonstrated by Jansen
et al. In a quantitative matter, 1H T1 NMR experiments are not in good agreement with
those found by Jansen et al. The interpretation of 1H T1 experiments is not always
straightforward, especially in this type of polymers where fast molecular dynamics of
rotational motions of (- CH3) groups and 180C flip rotation of benzene groups may
well contribute to spin relaxation, especially at high temperature where spin diffusion is
low. However static 1H line width analysis (deconvolution and integration) was used for
quantification and the results obtained for the rigid fractions are found to be in good
agreement with these found by Jansen et al. [7,10].
A higher mobility than the one found in CP experiments was depicted in 1H T1 NMR
experiments at 353 K. This seems to further confirm that Dianol is not only located in
the most mobile fraction but also in the amorphous interphase.
For quantification of phase morphology into the modified PBT samples, T1 analysis
seems to hold few ambiguities, especially as spin diffusion as well as local mobility
influence the relaxation decay. For such reason T2 relaxation solid-echo and Hahn-echo
analyses for a more accurate quantification of the different fractions were performed and
are presented in the following Chapter 4.
Chapter 3
57
3.6. References
[1] Backson, S. C. E.; Kenwright, A. M.; Richards, R. W. Polymer 1995, 36, 1991.
[2] Kim, J. H.; Lyoo, W. S.; Ha, W. S. J. Appl. Polym. Sci. 2001, 82,159.
[3] Marchese, P.; Celli, A.; Fiorini, M. J. Polym. Sci., Polym. Phys. 2004, 42, 2821.
[4] Fernandez-Berrediti, M. J.; Iruin, J .J.; Maiza, I. Polymer 1995, 36, 1357.
[5] Berti, C.; Colonna, M.; Fiorini, M.; Lorenzetti, C.; Marchese, P. Macromol. Mat. Eng. 2004, 289, 49.
[6] Kim, J. P.; Lyoo, W. S.; Ghim, H.D. Polymer 2003, 44, 895.
[7] Jansen, M. A. G.; Goossens, J. G. P.; de Wit, G.; Bailly, C.; Koning C. E. Macromolecules 2005, 38, 2659.
[8] Hait, S. B.; Sivaram, S. Macromol. Chem. Phys. 1998, 199, 2689.
[9] Kimura, M.; Porter, R. S. J. Polym. Sci., Polym. Phys. 1983, 21, 367.
[10] Jansen, M. A. G.; Goossens, J. G. P.; de Wit, G.; Bailly, C.; Koning C. E. Anal. Chim. Acta 2006, 557, 19.
[11] Packer, K. J; Pope, J. M.; Yeung, R. R.; Cudby, M. E. A. J. Polym. Sci. 1984, 22, 589.
[12] Havens, J. R.; Vanderhart, D .L. Macromolecules 1985, 18, 1663.
[13] Hentschel, D. R.; Sillescu, H.; Spiess, H. W. Macromolecules 1981, 14, 1605.
[14] Schmidt-Rohr, K.; Spiess, H. W. Multidimensional Solid-State NMR and polymers; Academic Press 1994.
[15] Miller, J. B. J. Thermal Analysis 1997, 49, 521.
[16] Yu, H.; Natansohn, A.; Singh, M. A.; Plivelic, T. Macromolecules 1999, 32, 7562.
[17] Hedesiu, C.; Demco, D. E.; Kleppinger, R.; Buda, A. A.; Blumich, B.; Remerie, K.; Litvinov, V. M. Polymer 2007, 48, 763.
[18] Yao, Y. F.; Graf, R.; Spiess, H. W. Macromolecules 2008, 41, 2514.
[19] Wind, M.; Brombacher, L.; Heuer, A.; Graf, R.; Spiess, H. W. Solid State NMR 2005, 27,132.
[20] Demco, D. E.; Litvinov V. M.; Rata, G.; Popescu, C.; Phan, K. H.; Schmidt, A.; Blumich, B. Macromol. Chem. Phys. 2007, 208, 2085.
[21] Hedesiu, C.; Demco, D. E.; Kleppinger, R.; Van der Poel, G.; Remerie, K.; Litvinov, V. M.; Blumich, B.; Steenbakkers, R. Macromol. Mater. Eng. 2008, 293, 847.
[22] Bertmer, M.; Wang, M.; Kruger, M.; Blumich, B.; Litvinov V. M. Chem. Mater. 2007, 19, 1089.
[23] Clauss, J.; Schmidt-Rohr, K.; Spiess, H. W. Acta Polym. 1993, 44, 1.
[24] Kenwright, A. M.; Say, B. J. Solid State NMR 1996, 7, 85.
Chapter 3
58
[25] Wang, J.; Jack, K. S.; Natansohn, A. J. Chem. Phys. 1997, 107, 1016.
[26] Magusin, P. C. M. M.; Mezari, B.; Van der Mee, L.; Palmans, A. R. A.; Meijer, E. W. Macromol. Symp. 2005, 230, 126.
[27] Adriaensens, P.; Storme, L.; Carleer, R.; D’Haen, J.; Gelan, J. Macromolecules 2002, 35, 135.
[28] Lequieu, W.; Van de Velde, P.; Du Prez, F. E.; Adriaensens, P.; Storme, L.; Gelan, J. Polymer 2004, 45, 7943.
[29] Goh, S. H.; Siow, K. S. Polym. Bull. 1987, 17, 453.
[30] Hong, J.; Goh, S. H.; Lee, S. H.; Siow, K. S. Polymer 1995, 36, 143.
[31] Hernandez, R.; Perez, E.; Mijangos, C.; Lopez, D. Polymer 2005, 46, 7066.
[32] Vanhaecht, B.; Willem, R.; Biesemans, M.; Goderis, B.; Basiura, M.; Magusin, P. C. M. M.; Dolbnya, I.; Koning, C. E. Macromolecules 2003.
[33] Bloembergen, E. M.; Purcell, R.; Pound, V. Physical Review 1948, 73, 679.
[34] Fedotov, V. D.; Schneider, H. Structure and dynamics of bulk polymers by NMR-methods, Springer 1989.
[35] Douglas, D. C.; Jones, G. P. J. Chem. Phys. 1966, 45, 956.
[36] Wang, J.; Cheung, M. K.; Mi, Y. L. Polymer 2001, 42, 3087.
[37] Wang, J.; Cheung, M. K.; Mi, Y. L. Polymer 2002, 43, 1357.
Chapter 4
59
Chapter 4
Phase composition of modified polyesters probed by solid-state 1H wideline NMR
of spin-spin relaxation
Summary
Solid-state NMR spectroscopy has been used to quantify morphological fractions of
poly(butylene terephthalate) (PBT)-copolymers obtained by the incorporation of 2,2-
bis[4-(2-hydroxyethoxy)phenyl]propane (Dianol) via solid-state copolymerization
(SSP). Different PBT / Dianol ratios have been investigated.
In this study, static solid- state 1H spin-spin relaxation experiments were used to
investigate the rigid fractions of the modified PBT samples. The results were fitted with
multi-exponential decay functions, or a combination of Gaussian and exponential
decays. Solid-Echo as Hahn-Echo pulse sequences have also been applied.
Chapter 4
60
4.1 Introduction
As already mentioned in Chapter 3, PBT is a semi-crystalline polymer, that presents a
broad range of application because of its ease of processability and high crystallization
rate. Nevertheless, the properties of the bulk polymer do not always meet the demand
and requirements, especially at elevated temperature. In order to broaden the range of
product applications, it might be useful to modify PBT in such a way that its high
crystallization rate is retained while its glass transition temperature (Tg) is increased. In
this chapter, modified PBT samples with a diol monomer obtained via solid-state
polymerization are investigated. The samples were synthesized in the solid-state by M.
Jansen et al., exhibing superior properties resulting from the modified microstructures
[1,2]. Jansen et al. also demonstrated that the trans-esterification reaction only occurs in
the amorphous phase of their samples modified in the solid state. Therefore, PBT blocks
are retained.
The diol was selected according to the following boundary conditions [1]:
- The diol should increase the Tg of PBT after incorporation.
- It should be commercially available.
- It should be thermally stable as the SSP reaction temperature is about 180 C to
210 C.
- The melting temperature (Tm) of the Diol should be lower then the one of the
PBT but higher than the glass transition (Tg) of PBT.
- For a trans-esterification reaction to occur between Diol and PBT, a good
miscibility condition of the Dianol is required.
- The diol should be highly reactive.
The 2,2-bis[4-(2-hydroxyethoxy)phenyl]propane (Dianol) was selected, that fulfilled all
the above requirements.
Jansen et al. demonstrated that Dianol was fully incorporated into the amorphous phase,
and that the microstructure of the modified PBT consists of a two-phase system
composed of a crystalline phase and an amorphous one, the latter subdivided into a rigid
amorphous fraction and a mobile amorphous one [2].
In Chapter 3, static solid-state NMR measurements of spin-lattice relaxation times in
the rotating frame T1 and T1 relaxation times were used to estimate 1H spin diffusion
Chapter 4
61
distances and to investigate the morphology of the modified PBT samples, together with
T1(1H) -filtered 1H-13C cross-polarization experiments (CP-MAS). From T1 (1H) and
T1(1H) measurements, an upper and a lower limit of domain sizes were established.
Furthermore, the presence of a three phase system was demonstrated, assigning a
crystalline phase, a rigid amorphous or interfacial phase and a mobile amorphous phase.
The results were found to be in reasonable agreement with Jansen et al. However, a few
discrepancies were found in phase’s quantification, which might be attributed to the use
of different techniques that probe different types of information.
The spin-lattice relaxation T1 processes are determined as well by molecular motions as
by spin-diffusion and they have been proved to be successful to study phase
morphology in many polymeric systems [3-11]. However, for highly crystalline rigid
polymers, the strong static proton dipolar interaction makes the spin-spin relaxation
process dominant in the relaxation. During the very short T2 relaxation process spin-
diffusion cannot happen. For this reason, T2 spin-spin relaxation measurements are
preferred when investigating crystalline and interfacial rigid fractions [12-15].
In this chapter, the results of static T2 (1H) relaxation experiments on the modified PBT
samples are discussed. The data of the transverse relaxation decay times were fitted with
two and three exponential decay models; the relaxation time values and the fractions of
mobile and rigid protons in each phase were extracted. The results have been analyzed
and compared to those calculated by Jansen et al. using modulated DSC (TMDSC). The
results are also compared to the values found by spin lattice relaxation in the rotating
frame as well as to the simulated fractions calculated starting from the 1H NMR
fractions obtained for the pure PBT sample.
Additional T2 (1H) relaxation experiments with a solid-echo detection sequence were
carried out on samples (BD96Di04)ssp and (BD54Di46)ssp at room temperature (RT), 353 K
and 435 K. As an extra reference, some T2 experiments with a Hahn-echo detection
sequence were carried out on the (BD54Di46)ssp sample. Mostly the best fit was found as
a combination of Gaussian and exponential (Lorentzian) decay. In general the values of
crystalline fraction were found in reasonable agreement with those found by Jansen et
al. and these obtained via T1 experiments (see Chapter 3). For the sample
(BD96Di04)ssp the high rigidity of the sample (highest Tg) however makes the
quantification of the crystalline phase challenging, even at 435 K well above Tg.
Chapter 4
62
4.2. Experimental Section
4.2.1. Sample preparation
The modified PBT samples were obtained and synthesized by M. Jansen [1].
The synthesis of PBT-Dianol copolymers by SSP is described in Jansen’s thesis.
Different ratios of PBT and Dianol were used and are reported in Chapter 3. BDx
corresponds to the fraction in mol% of PBT (expressed in 1,4-butanediol units) whereas
Diy corresponds to the fraction in mol% of Dianol in the initial mixture used for the
reaction (BDxDiy)feed. After SSP, the initial mol fractions slightly differ from
(BDxDiy)ssp. This difference may be attributed to the evaporation of 1,4-butanediol or
ethylene glycol during the SSP reaction.
4.2.2. NMR experiments
The NMR experiments were performed in Eindhoven (TU/e) on a DMX500 Bruker
spectrometer operating at 500 MHz for protons, and a 7-mm probe head was used. The
relaxation experiments were carried out at 435 K using a variable temperature unit. For
such high temperature, calibration was performed using the melting point of organic
compounds. The samples were equilibrated at each temperature for 20 minutes. The
Hahn-Echo Pulse Sequence (HEPS) (see below) was used to record the decay of the
transverse magnetization Mz at first. Later at the U-Hasselt a Solid-Echo Pulse
Sequence (SEPS) developed by Powles and Strange was also applied [16].
4.2.2.1. Hahn-Echo pulse (HEPS), T2 relaxation experiments
A Hahn-Echo (HEPS) measurement is the production of spin echo by radiofrequency
(rf) pulses. The HEPS, consists of a 900 radiofrequency pulse which flips longitudinal
magnetisation Mz into the xy’-plane, whereby the transverse magnetization Mxy’ starts to
precess with the Larmor frequency [17]. After half the echo time τ, the preparation
phase is followed by a refocusing 1800 pulse which serves to generate an echo signal.
The sequence is based on a double-pulse: 900 - τ -1800
- τ – acquisition (Figure 4.1.a).
The echo serves to re-phase all the coherences, which are lost during the time τ between
900 and 1800 pulses due to dephasing along the y’-axis. This dephasing is due to
Chapter 4
63
extrinsic static magnetic field inhomogeneities (i.e. field inhomogeneities of the main
magnetic field) and intrinsic dephasing coming from the examined material.
The real echo-time (TE) is half of the 900 pulse + 2τ + the 1800 pulse. In the following
experiments, the time τ (between two pulses) is varied from 13 s to 170 ms for samples
(BD54Di46)ssp and (BD76Di24)ssp, from 10 s to 4 ms for sample (BD96Di04)ssp and from
13 s to 4 ms for sample (BD100Di00)ssp. The second pulse in the HEPS inverts nuclear
spins and an echo signal is obtained with a maximum at time t = TE after the first pulse.
By varying the pulse spacing, intensity profiles are obtained as function of time τ. The 1H-NMR T2 HEPS relaxation signal is acquired as a function of echo time t starting
from the top of the echo. The experimental data were fitted with a multiple-exponential
decay fit (or Lorentzian L) using multivariate analysis:
lis TtCTtBTtAtM expexpexp)( (4.1)
Where M(t) represents the magnetization function, A is proportional to the fraction of
components with a short relaxation time, B to the fraction of components with an
intermediate relaxation time and C to the fraction of components with long relaxation
time; t = TE is the real echo-time in the HEPS experiment.
Figure 4.1.a. Hahn-Echo pulse sequence for a weakly coupled system, after a single quantum refocus.Solid echo proton wide-line (static) NMR measurements (SEPS) were carried out at ambient temperature, 353 K and 435 K on a Varian Inova 400 spectrometer using a dedicated wide-line probe equipped with a 5 mm coil.
4.2.2.2. Solid echo experiments (SEPS), T2 relaxation experiments
Solid echo proton wide-line (static) NMR measurements were carried out at ambient
temperature, 353 K and 435 K on a Varian Inova 400 spectrometer in a dedicated
wide-line probe equipped with a 5 mm coil.
Chapter 4
64
On-resonance Free Induction Decays (FIDs) were acquired by applying the solid-echo
pulse technique (900x’-tse-900
y’-tse-acquisition) in an effort to overcome the effect of
dead-time of the receiver [5]. The 900 pulse length t90 was set to 1.5 s and spectra were
recorded with a spectral width of 2 MHz (0.5 s dwell time) allowing an accurate
determination of the echo maximum. The solid echo is formed with a maximum at t =
(3t90/2 + 2tse). A preparation delay of 5 times the T1H relaxation decay time was always
respected between successive accumulations to obtain quantitative results.
For the analysis of the T2H-FIDs combinations of Gaussian (G) and Lorentzian
(exponential) (L) fitting methods have been used (see equations 4.5.1 and 4.5.2). For the
two-component analyses a GL-combination (equation 4.5.1) showed a reasonable result
for solid (G) and mobile (L) phases. For the three-component method the GGL-
combination showed the best acceptable result for solid-crystalline (G), solid-
amorphous (G) and mobile-amorphous (L) phases.
For the (BD54Di46)ssp sample at 435 K the Solid-Echo detection method (SEPS), using
the analysis starting at the maximum FID-intensity, produced extra FID-oscillations
because the chemical shifts of the proton signals of the most mobile phase resolved in 3
peaks at 400 MHz (4 peaks at 500 MHz). Therefore, an adapted Solid-Echo experiment
with variable echo-time in the detection was implemented. For the analysis, the total
peak area after FT was used as a function of the variable echo-time. The solid-echo
pulse sequence is shown in Figure 4.1.b.
Figure 4.1.b. Solid echo pulse sequence for a strongly coupled system, after a double quantum refocus.
As an alternative, Hahn-Echo detection pulse sequence (HEPS) has also been carried
out, using the same dedicated wide-line probe.
Chapter 4
65
4.3. Results and discussion
4.3.1. T2 spin-spin relaxation experiments of Hahn-Echo pulse sequence (HEPS)
It is well known that in the solid state the proton NMR spectrum of semi-crystalline
polymers at temperatures between the glass transition and the crystal melting
temperature is characterized by superimposed broad and narrow resonances. Protons, in
the rigid phase, give rise to dipolar broadened resonances while protons in the
amorphous phase give rise to motional narrowed resonances. As already mentioned in
the introduction, several studies have been done to determine the crystallinity, phase
composition and mobility of semi-crystalline polymers [18-29].
Figure 4.2.a. T2 Relaxation decays curves recorded at 435 K (experimental data fitted with three exponential decay model ). () correspond to sample (BD100Di00)ssp; () correspond to sample (BD96Di04)ssp; ( ) correspond to sample (BD76Di24)ssp, () correspond to sample (BD54Di46)ssp.
Chapter 4
66
In this study, 1H – NMR spin-spin Hahn-Echo relaxation experiments (HEPS) have
been applied to the modified PBT samples. The transverse magnetization decays of
these samples above the glass transition and below the melting temperature were fitted
with a multi-component exponential decay fit (LLL).
For all samples measured at 500 MHz and 435 K, the FID can be decomposed into three
components. The data were best fitted with a three-exponential decay model (Figure
4.2.a) suggesting three types of 1H mobilities in the samples. This can correspond to
mobility of 1H-atoms contained in the crystalline phase (very fast decay, T2s), to the
rigid amorphous fraction (intermediate decay T2i) and to the mobile amorphous fraction
(slow decay T2l) (Table 4.1.a and 4.1.b).
Like the FID, the spectra could be deconvoluted into three components. Broad,
intermediate and narrow lines originate from these three components, the faster
decaying component showing the broader line.
Figure 4.2.b. Relaxation decays curves recorded at 435K (experimental data fitted with two exponentials decay model). () correspond to sample (BD100Di00)ssp; () correspond to sample (BD96Di04)ssp; () correspond to sample (BD76Di24)ssp, () correspond to sample (BD54Di46)ssp.
Chapter 4
67
A fitting with a two-component decay could only be accepted for the (BD54Di46)ssp
sample at 435 K as observed in Figure 4.2.b and Table 4.1.c.
Table 4.1.a. T2 relaxation times obtained with three exponential decay fit (2D analysis) at 435 K.
FBD-T, total FDi-T, total Crystalline T2s (μs)
Rigid amorphous Mobile amorphous [mol%] [mol%] T2i (μs) T2l (ms)
100 0 10 120 1.4
96 4 11 70 0.27
76 24 10 300 1.2
54 46 7000 10.000 41
Table 4.1.b. T2 relaxation times obtained with three exponential decay fit (1D analysis) at 435 K.
FBD-T, total FDi-T, total Crystalline T2s (μs)
Rigid amorphous Mobile amorphous [mol%] [mol%] T2i (μs) T2l (ms)
100 0 16 156 1.08
96 4 9 72 0. 28
76 24 43 391 1.5
54 46 62 1998 9
Table 4.1.c T2 relaxation times obtained with two exponential decay fit (2D analysis) at 435 K.
FBD-T, total
[mol%] FDi-T, total T2l (ms) T2s (s)
Rigid [mol%] Amorphous
100 0 14 0.60
96 4 12 0.17
76 24 80
0.8
54 46 4340 10
As already shown in Chapter 3, the quantitative analysis for the FIDs and wide-line
NMR spectra (after FT) in terms of crystalline, rigid amorphous and mobile amorphous
can be complicated due to a number of reasons.
Chapter 4
68
First, one should perform the experiments at a temperature well above the Tg to probe
more mobility in the sample in order to be able to distinguish different phases. However,
this requirement implies that some phase rearrangements may occur. So the temperature
should be high enough but not too high to prevent annealing effects. This has been
checked by MDSC (results presented in Chapter 3) but also in NMR experiments.
Values of relaxation times and corresponding fractions were found to remain the same
after temperature treatment (heating to 435 K and subsequent cooling).
The T2 relaxation times at 435 K are shown in Table 4.1. The data were fitted with 1D
and 2D methods. The results show that, when more Dianol is incorporated, the 1D and
2D data diverge. As already observed in Chapter 3, the 2D analyses do not seem to
reflect the phase behavior when more Dianol is incorporated. The fact that the 2D
analysis takes the chemical shift into account may not always reflect the phase
relaxation (see Appendix 1 & 2). It may also assimilate relaxation of protons in a phase
such as side groups (methyl group: Me-) to the overall protons relaxation in that phase
relaxation. The 1D data seem to better describe the physical reality. However, a 2D
analysis seems to help when trying to localize the Dianol (see Chapter 3).
For all samples, except for (BD54Di46)ssp, the shorter 1H relaxation times obtained
through 1D and 2D are in the order of a few s, which is the typical order of magnitude
of T2 relaxation of 1H-atoms present in a crystalline phase where molecular motions are
restricted and relaxation decay is very fast as seen in Figure 4.2.a. These samples,
except for (BD54Di46)ssp, also show 1H mobilities with T2-values in the order of 100 s
which could be assigned to 1H in an intermediate rigid phase, corresponding to the rigid
amorphous fraction as mentioned by Jansen et al. [1,2]. The third type of mobility is
attributed to 1H-atoms in the most mobile phase (amorphous phase) and shows a
relaxation time in the order of the ms.
For the sample (BD54Di46)ssp with the highest mole fraction of Dianol, the T2 values
obtained with a 2D fitting, are much higher, the three relaxation times are in the same
order of magnitude (tenths of ms) and, at first sight, a two component fit (LL) seems to
be more realistic. These data should be treated with care, and as mentioned above, for
this sample, a 1D fitting is more realistic.
In highly crystalline polymers, the strong static dipolar interactions make the spin-spin
relaxation process more efficient. In the case of highly amorphous systems such as
Chapter 4
69
(BD54Di46)ssp, the spin-spin relaxation process at 435 K is less efficient because of the
faster fluctuating dipole fields caused by conformational motions, narrowing the
resonance lines. The mobility in sample (BD54Di46)ssp is very high especially at such
experimental conditions. The amorphous phase in the sample behaves almost as a liquid.
Too much isotropization is present to be able to distinguish between different
amorphous fractions. Figure 3.8 (see Chapter 3) shows a rather high resolution (as in
liquid like), and we could distinguish four narrow peaks corresponding to proton
resonance frequencies of aromatic (2 peaks), ester, and aliphatic 1H-atoms. In addition,
in sample (BD54Di46)ssp, the methyl group -(Me) content is relatively high as compared
to other samples. Local mobility rather then overall phase mobility may be depicted,
which results in a challenging quantification of the amorphous phase for this sample.
From a bi-exponential (LL) fit, one can determine the fraction of rigid protons
associated with the decay of the broad components (fast decay) and the fraction of more
mobile protons, associated with the decay of the narrow component. The fractions of
each component are reported in Tables 4.2.a, 4.2.b (three-phase fit) and 4.2.c (two-phase
fit).
One should note that the crystalline fraction and rigid amorphous fractions calculated by
Jansen et al. via TMDSC and reported in Chapter 3, are based on mol fractions. The
crystalline, rigid amorphous, and mobile amorphous fractions were determined by
Jansen et al., based using the following equations [1,2]:
mobile= cp / cp 0 (4.2)
heating= H melting/H 0fusion (4.3)
rigid= 1- heating + mobile (4.4)
where cp is the heat capacity increase at the half- step Tg of the non-transesterified
PBT-Dianol mixture, whereas cp0 is the heat capacity increase for 100% amorphous
PBT at the half step Tg ; H0fusion is the enthalpy of 100% crystalline PBT. The value of
cp0 for 100% amorphous PBT was assumed to be independent of the added Dianol
monomer.
Chapter 4
70
Table 4.2.a. Rigid fraction, intermediate fraction and mobile fraction obtained from static T2 experiments using a three exponentials decay fit (2D analysis).
FBD-T, total [mol%]
FDi-T, total
[mol%] Crystalline
(%) Rigid amorphous
(%) Mobile amorphous
(%)
100 0 51 22 27
96 4 74 15 11
76 24 36 31 33
54 46 22 72 2
Table 4.2.b. Rigid fraction, intermediate fraction and mobile fraction obtained from static T2 experiments using a three exponentials decay fit (1D analysis).
FBD-T, total [mol%]
FDi-T, total
[mol%] Crystalline
(%) Rigid amorphous
(%) Mobile amorphous
(%)
100 0 46 24 30
96 4 63 17 20
76 24 7 55 37
54 46 0.1 3 97
Table 4.2.c. Rigid fraction and amorphous fraction obtained from static T2 experiments using a two exponentials decay fit (2 D analysis).
FBD-T, total [mol%]
FDi-T, total [mol%]
Rigid fraction (%)
Amorphous fraction (%)
100 0 82 18
96 4 78 22 76 24 46 54
54 46 11 89
When looking at values presented in Tables 4.2, the fractions of rigid protons, and
mobile protons obtained by fitting the experimental data with the two exponential (LL)
decays (Table 4.2.c) seem to be in reasonable agreement with the summation of rigid
and amorphous fractions found by Jansen et al., also in good agreements with values
found via peak deconvolution of T1 data at short evolution (spin-lock) time presented in
Chapter 3 (Table 3.7).
For the fractions obtained using a three components fit on the T2-data from the Hahn-
echo experiments, a bad correlation has been found with Jansen’s fractions especially
Chapter 4
71
for the (BD54Di46)ssp sample. For the (BD96Di04)ssp sample the value of rigid (crystalline)
fraction is found to be quite high, probably due to the fact that this sample is much more
rigid showing a much higher Tg. Also, the intermediate (rigid amorphous) fraction is
expected to be more important in this sample; so some part of this intermediate fraction
may be seen in T2 spin-spin relaxation experiments as part of the rigid (crystalline) one.
Furthermore, the T2 values are more or less in agreement with those calculated from
pure PBT while assuming that Dianol is incorporated in all the amorphous phase. The
values are reported in Table 4.3.
Table 4.3 Percentage of the two 1H fractions present in the system obtained via calculation method based on pure PBT and assuming Dianol is incorporated in the totality of both amorphous phases.
FBD-T, total [mol%]
FDi-T, total [mol%]
Rigid fraction (%)
Amorphous fraction (%)
100 0 51 49
96 4 48 52 76 24 35 63
54 46 28 72
One should note that phase quantification has to be treated with real care.
On one hand, different methods may be used for determination of crystallinity but these
do not necessarily lead to the same crystalline fractions. Some techniques look at 1H
crystallinity such as NMR, while DSC for instance looks at crystal melts.
During the fast T2 relaxation in the solid state NMR experiments, spin diffusion cannot
happen. Therefore the choice of T2 was made to better quantify the phase fractions.
Even then the distinction between the rigid amorphous and the mobile amorphous
phases is not always straightforward. When performing T2 experiments at higher
temperature, the increased local molecular dynamics can cause increased spectroscopic
resolution, increasing the number of mobile components in the analysis. This means that
the mobile amorphous phase will show up in different components in the multi-
exponential analysis of the FID. The number of components does not reflect anymore
the number of phases present in the system. However because T2 is always decreasing
Chapter 4
72
with decreased molecular dynamics, the phase assignment to corresponding T2 values is
still more straightforward compared to T1 and T1.
The values of the total rigid fraction (sum of the crystalline and rigid amorphous phases)
obtained for the pure PBT fractions from T1, T2 (2D, 1D) and from Jansen are all in
reasonable agreements. It is only when incorporating the Dianol that the relaxation
picture starts to become more complex and ambiguous. Effectively, at higher
temperature (above Tg) Dianol contains highly mobile groups (CH3- and p-phenyl
groups). When Dianol is incorporated in the amorphous phases, these mobile groups
generally cause an increase of the T2H relaxation time.
At low concentration (BD96Di04)ssp the incorporation of Dianol induces a minor
reorganization in the amorphous phases, slightly reducing the molecular dynamics,
resulting in a decrease of their T2-values (Tables 4.1). At higher Dianol concentration
(BD76Di24)ssp , a significant increase of the T2-values of the amorphous phases is noticed
without a change of the T2-value of the crystalline phase. For the (BD54Di46)ssp sample,
even the crystalline PBT-phase is influenced by the Dianol incorporation, showing a
significant increase in molecular dynamics.
Another source of errors in the phase quantification can also be found in the use of a
CP/MAS probe for the static solid-state NMR measurements. Such a CP/MAS probe
has a rather long dead time, only allowing proper phase quantification based on T2H-
values longer than about 20 s.
For a better understanding and to lift ambiguity in quantification of the samples
(BD96Di04)ssp with the least amount of Dianol and the sample (BD54Di46)ssp with the
higher amount of Dianol, some complementary measurements have been performed on
a Varian Inova 400 spectrometer using a dedicated wide-line probe equipped with a 5
mm coil. Thanks to the shorter dead time of this wide-line probe, solid-state relaxation
experiments with Solid-Echo (SE) detection can be performed. T1, T1 (Chapter 3) and
T2H measurements using such a solid-echo detection have been performed on samples
(BD96Di04)ssp and (BD54Di46)ssp at 293 K, 353 K and at 435 K.
4.3.2. Analysis and relaxation measurements for sample (BD96Di04)ssp
For the (BD96Di04)ssp sample at room temperature, the experimental T2 data were best
fitted with a single Gaussian function. At 353 K a bi-component (GL) best fit, using a
Chapter 4
73
combination of Gaussian and exponential (Lorentzian) functions as detailed in equation
(4.5.1) results in an amount around 93% for the rigid phase and around 7% for the
mobile phase.
At 435 K, both a tri-component (GGL)-fit (equation 4.5.2) and a two-component (GL)-
fit have been performed (equation 4.5.1):
M (t) = A exp (-1/2(t/T2s ) 2) + B exp (-t/Tl ) (4.5.1)
M (t) = A exp (-1/2(t/T2s )2) + B exp (-1/2(t/T2i )
2) + C exp (-t/T2l ) (4.5.2)
The (GL)-fit as well as the (GGL)-fit result in an amount around 70% for a rigid phase
and around 30% for an amorphous phase. From the tri-component fit, this amorphous
phase seems to be composed of 17 % of the solid-amorphous inter-phase and 12 % of
the mobile-amorphous phase.
The T2H-values and the fractions of the Solid-Echo experiments at 400 MHz (Table 4.4)
fit rather well with the values from corresponding Hahn-Echo experiments at 500 MHz
(Tables 4.1 and Tables 4.2).
Table 4.4. Data for (BD96Di04)ssp extracted from a combination of Gaussian (G) and exponential (L) functions from solid-echo experiments. At RT (20 °C) only a Gaussian fit was used.
(BD96Di04)ssp Crystalline
T2s (μs) / f (%)
Rigid-Amorphous
T2 i (μs) / f (%)
Mobile-amorphous
T2l (μs) / f(%)
(T = 20 °C); G 11.9 (one phase) - -
(T= 20 °C); G after 162°C 12.1 (one phase) - -
T=353 K (80 °C) GL 13,8 (93.4) 22,3 (6.6) -
T=435 K (162 °C) GGL 16.3 (71.2) 45 (16.8) 114 (12.0)
T=435 K (162 °C) GL 16,3 (71.5) - 117 (28.5)
4.3.3. Analysis and relaxation measurements for sample (BD54Di46)ssp
For the sample (BD54Di46)ssp a Solid-Echo pulse sequence as well as a Hahn-Echo pulse
sequence were applied. One should note that phase quantification in sample
(BD54Di46)ssp may be ambiguous, as the amount of Dianol is much higher than in the
other samples, causing a much higher molecular mobility. As discussed in Chapter 3 of
Chapter 4
74
this thesis, this correlates with the low Tg (51°C) of this sample. The excess Dianol
leads on one hand to a partial dissolution of the PBT crystals and on the other hand to
inner-outer trans-esterification reactions as describes by Jansen et al. that imply a lower
average chains Mw for this copolymer [1,2].
4.3.3.1. T2H analysis with Solid-Echo detection
Different combinations of Gaussians (G) and exponentials (L) were used to fit the
experimental data for sample (BD54Di46)ssp. The results are reported in Table 4.5.a.
At room temperature (before and after cooling down from 435 K) a GL two-component
fit resulted in a rigid phase of about 80 % and a more mobile phase of about 20 %.
However the T2-values of the rigid and mobile phase are rather close to each other (13
and 20 s) implying that their molecular mobilities are similar and very slow. Being
below Tg an important part of the rigid amorphous phase is expected to be integrated in
the crystalline phase.
Table 4.5.a. Data for (BD54Di46)ssp extracted from a combination of Gaussian and exponential functions from Solid-Echo experiments (1 D analysis).
(BD54Di46)ssp Crystalline
T2s (μs) / f (%) Rigid-amorphous
T2 i (μs) / f (%) Mobile-amorphous
T2 l (μs) / f (%)
(20°C)-GL 13 (81,8) 20 (18,2)
(20°C)-after162°C-GL 13 (81,1) 24 (18,9)
T=353 K (80 °C) -GL 16 (62,4) 20 (37,6)
T=435 K (162 °C) -GL 15 (21,8) 154 (78,2)
T=435 K (162 °C) -GL-f()
104 (58,1) 3305 (41,9)
T=435 K (162 °C) -GGL-f()
26 (10,9) 127 (51,4) 3812 (37,7)
T =435 K (162 °C) -GGL-f()- Abragam
33 (14,3) 291 (47,2) 3703 (38,4)
At 353 K, slightly above Tg, a similar GL two-component fit results in comparable T2H-
values but with an increase in the more mobile fraction (from 18 to 38 %).
At 435 K a difficult GL-analysis, based on a reduced set of data points, ends up with a
fraction of 22 % for the solid phase and a T2-value around 15 s, a value typical for a
crystalline phase at temperatures sufficient above Tg. The amorphous fraction of 78 %
Chapter 4
75
has a T2H-value of 154 s showing a significantly higher molecular mobility. However
this two-component analysis produces an insufficient accurate fit. A three-component
analysis on the same single FID at this higher temperature, produced by a routine SE-
experiment, is not fitting at all. At 435 K an important part of the amorphous phase of
this (BD54Di46)ssp sample has an increased molecular mobility resulting in a
spectroscopic resolution of the chemical groups in this phase. After FT, three peaks are
resolved at 400 MHz corresponding to aromatic, ether and CH3- + CH2R groups. This
increased spectroscopic resolution causes extra oscillations on the FID-signal, which
complicates the multi-component fit of this FID. Normally in the SE-experiment, the
analysis of the FID-signal is done based on a single echo-time. The extra oscillations,
superimposed on the normal FID, make a proper multi-component analysis impossible.
A simple adaptation of the SE pulse-sequence allows a T2-determination at variable
echo-times. This allows signal area measurements, after FT and proper peak phasing, as
a function of the echo-time, avoiding the problem of chemical shift oscillations. Based
on such an adapted SE-experiment at 435 K, we performed a T2 two-component GL-
analysis and a T2 three-component GGL-analysis using the total peak area as a function
of the echo-time. For the two-component fit T2-values of 104 and 3305 s, for a rigid
and a mobile phase with 58 % and 42 % fractions are found (Table 4.5.a). The three-
component GGL-analysis however clearly produced a better fit (Figure 4.3). T2-values
of 26, 127 and 3812 s, corresponding respectively to a crystalline, a rigid amorphous
and a mobile amorphous phase with 11 %, 51 % and 38 % fractions are found (Table
4.5.1). This lower crystalline fraction is already closer to what is expected. A refinement
of the GGL-analysis can be done by introduction of the Abragam algorithm for the
second Gaussian function in the tri-component fit (see equation 4.6) [29,30]:
M (t) = A exp (-1/2(t/T2i ) 2) + B exp (-1/2(t/T2s )
2).(sin(2(t/))/ (2(t/)))
+ C exp (-t/T2l) (4.6)
The parameter introduces an oscillation, which in this case could be linked to rather
fast Me- and p-phenyl group rotations [31]. Such an analysis gives T2-values of 33, 291
and 3703 s, for the crystalline, rigid amorphous and mobile amorphous phase with 14
%, 47 % and 39 % fractions respectively (Table 4.5.a).
Chapter 4
76
It is known that a ‘90° -- 90° -- acquisition’ SE-sequence can lead to an incomplete
refocus after the echo-time, especially with samples which hold mobile phases [4]. The
second 90°-pulse does not properly refocus e.g. chemical shift differences and p-phenyl
and methyl rotations.
Figure 4.3. T2 relaxation decays curves recorded on a 400 MHz wide-line probe at 435 K with a Solid-Echo sequence for sample (BD54Di46)ssp (experimental data fitted with a tri-component GGL-fit (thick line corresponds to an Abragam fit) and a two component GL-fit (thin line)).
4.3.3.2. T2H with Hahn-Echo detection
As a comparison Hahn-Echo (HE) experiments, having a better refocusing -pulse, are
performed on the same sample at 435 K. As well chemical shift artifacts as rotation
phenomena are refocused in this case. Thanks to the shorter dead time of the dedicated
wideline probe, used on the 400 MHz system, the shorter T2-values of the very rigid
phase can be measured more accurately (more data points at short echo-times, compare
Figure 4.2.a with Figure 4.4). Also, the total peak area and the area of the individual
Chapter 4
77
peaks are used for a tri-component fit. For the resolved peaks a GGL-combination,
using equation (4.7), leads to the most optimal fit (Figure 4.4):
M (t) = A exp (-1/2(t/T2s )2) + B exp (-1/2(t/T2i )
2) + C exp (-t/T2l) (4.7)
An analysis of the total peak area resulted in T2H-values around 20, 400 and 4000 s for
the crystalline, rigid-amorphous and mobile-amorphous phases. They correspond with
the individual fractions of 26 %, 17 % and 57 %. (Table 4.5.b).
Table 4.5.b. Data for (BD54Di46)ssp extracted from a combination of Gaussian and exponential functions from Hahn-Echo experiments recorded at 435 K(1 D analysis).
Crystalline Rigid amorphous T2 i (μs)/ f (%)
Mobile amorphous T2l (μs)/ f (%)
(BD54Di46)ssp T2 s (μs) / f (%)
T =435 K (162 °C) -Total GGL
20 (25,8) 370 (17.2) 4254 (57.0)
Figure 4.4. T2 relaxation decays curves recorded on a 400 MHz wide-line probe at 435 K with a Hahn-Echo sequence for sample (BD54Di46)ssp (experimental data fitted with a tri-component GGL-fit).
Chapter 4
78
The small fractions of the crystalline phase (11 % from the SE-experiment and 26 %
from the HE-experiment) are representative for the reduction of the PBT-crystallinity
caused by the SSP-reaction with 46 % of Dianol. This is in good agreements with the
low crystalline fraction found by Jansen et al. via MDSC as well as with the rigid
fraction values found from tri- and two-component fits in Table 4.2.a (22 %), Table
4.2.b (0.1 %), and Table 4.2.c (11%) measured at 500 MHz.
4.4. Conclusion
Solid-state 1H T2 relaxation measurements allow a more straightforward analysis for the
different phases of the (BDxDiy) copolymers. As already observed in Chapter 3, the
results show a crystalline fraction that decreases with Dianol concentration.
In the rather rigid sample (BD96Di04)ssp, the crystalline phase may be over-estimated
because part of the rigid amorphous fraction (or interface) may be depicted in the
crystalline phase.
For the (BD54Di46)ssp sample, the crystalline PBT-phase is quite influenced by the Dianol
incorporation, showing a significant increase in the molecular dynamics. The pure PBT
sample shows results similar to those obtained with other techniques indicating the
presence of a “three phases” system, composed of a really rigid phase (crystalline
fraction), an intermediate phase (interfacial fraction) and a mobile amorphous phase.
When Dianol is incorporated, the three phases system holds, but the relaxation picture
and phases quantification becomes challenging, especially with this type of Diol
containing high mobile groups at temperatures high enough (well above Tg), causing
chemical shift resolution. This increased spectroscopic resolution may cause extra
oscillations on the FID-signal as observed in sample (BD54Di46)ssp, which complicates
the multi-component fit of the FID. At such high concentration of Dianol, only a
traditional 1D data analysis is realistic and the best fit was found using a combination of
Gaussian and exponential decays (GGL). The results obtained from Hahn-echo and
Solid-echo experiments were found to be in good agreements.
Chapter 4
79
4.5. Reference
[1] Jansen, M. A. G.; Goossens, J. G. P.; de Wit, G.; Bailly, C.; Koning, C. E. Macromolecules 2005, 38, 2659.
[2] Jansen, M. A. G.; Goossens, J. G. P.; de Wit, G.; Bailly, C.; Koning, C. E. Anal. Chim. Acta 2006, 557, 19.
[3] Packer, K. J; Pope, J. M.; Yeung, R. R.; Cudby, M. E. A. J. Polym. Sci. 1984, 22, 589.
[4] Hedesiu, C.; Demco, D. E.; Kleppinger, R.; Buda, A. A.; Blumich B.; Remerie, K.; Litvinov, V. M. Polymer 2007, 48, 763.
[5] Lequieu, W.; Van De Velde, P.; Du Prez, F. E.; Adriaensens, P.; Storme, L.; Gelan, J.; Polymer 2004, 45, 7943.
[6] Schmidt-Rohr, K.; Spiess, H. W. Multidimensional Solid-State NMR and Polymers, Academic Press, 1994.
[7] Miller, J. B. J. Thermal Analysis 1997, 49, 521.
[8] Yu, H.; Natansohn, A.; Singh, M. A.; Plivelic, T. Macromolecules 1999, 32, 7562.
[9] Havens, J. R.; Vanderhart, D. L. Macromolecules 1985, 18, 1663.
[10] Fedotov, V. D.; Schneider, H.; Structure and dynamics of bulk polymers by NMR-methods, Springer, 1989.
[11] Yao, Y.-F.; Graf, R.; Spiess, H. W. Macromolecules 2008, 41, 2514.
[12] Demco, D. E.; Litvinov V. M.; Rata, G.; Popescu, C.; Phan, K-H.; Schmidt, A.; Blumich, B. Macromol. Chem. Phys. 2007, 208, 2085.
[13] Hedesiu, C.; Demco, D. E.; Kleppinger, R.; Poel, G. V.; Remerie, K.; Litvinov, V. M.; Blumich, B.; Steenbakkers, R. Macromol. Mater. Eng. 2008, 293, 847.
[14] Bertmer, M.; Wang, M.; Kruger, M.; Blumich, B.; Litvinov, V. M. Chem. Mater. 2007, 19, 1089.
[15] Hentschel, D. R.; Sillescu, H.; Spiess, H. W. Macromolecules 1981, 14, 1605.
[16] Powles, J. G.; Strange, J. H. Proc. Phys. Soc. London 1963, 82, 6.
[17] Bloch, F.; Hansen, W. W.; Packard, M. Phys. Rev. 1946, 69, 127.
[18] Tanaka, H.; Kohrogi, F.; Suzuki, K. Eur .Polym. J. 1989, 25, 449.
[19] Bergmann, K.; Nawotki, K. Z. Polymer 1967, 219, 132.
[20] Earl, W.; Vanderhart, D. L. Macromolecules 1979, 12, 762.
[21] Dadayly, D.; Harris, R. K.; Kenwright, A. M.; Say, B. J.; Sunnectioglu, M. M. IRC Polym. Technol. 1994, 35, 4083.
[22] Tanaka, H.; Inoue,Y. Polym. Int. 1993, 31, 9.
[23] Henning, J. Concepts in Magn. Res. 1991, 3, 125.
[25] Litvinov, V. M.; Penning, J. P. Molecular Chem. Phy. 2004, 205, 1721.
Chapter 4
80
[26] Schneider, M; Gaspar, L.; Demco, D. E.; Blumich, B. J. Chem. Phys. 1999, 111, 402.
[27] Tanaka, H.; Kohrogi, F.; Suzuki, K. Eur. Polym. J. 1989, 25, 449.
[28] Kitamaru, R.; Horii, F. Adv. Polym. Sci. 1978, 26, 139.
[29] Abragam, A.; The principle of NMR. Oxford University Press, 1991.
[30] Abragam, A.; Goldman, M.; NMR: order and disorder, Oxford University Press, 1982 (1).
[31] Sergeev, N. A.; Olsezewski, M. Solid State NMR 2008, 34, 162.
Chapter 5
81
Chapter 5
Heterogeneous chain dynamics in polyester network probed by 1H static spin-spin relaxation NMR experiments
Summary
Characterization of polymer networks is a very complex task, which has captivated the
attention of polymer scientists, especially during the last two decades. To be able to
improve and optimize the macroscopic properties of cross-linked resins, we must
develop a thorough understanding at the microscopic level, including features such as
the cross-link density of a network, the number of dangling chains, loops, and eventual
polymer chains unattached to the network (commonly called the sol fraction). The
chemical structure (topology) of cross-linked networks is complicated. During the
formation of a network, many different structures may be formed, resulting in a very
heterogeneous network. Different initiation propagation termination and combination
reactions may occur, as well as side reactions with other components present in the
curing resin system. Many parameters affect the curing process, including the
temperature, the solvent, the chemical and physical properties of the (pre)-polymers,
additives and catalysts presents. The final network structure with all its variations,
heterogeneities and defects determines the properties of the network.
In this study heterogeneous chain dynamics in a cured unsaturated polyester resin (UPR-
Palatal intermediate) are investigated using 1H-NMR transverse relaxation experiments.
Different transverse 1H relaxation times T2 are reported at different temperatures. T2
measurements were performed on a cross-linked sample in the original and swollen state
at respectively 4.7 and 11.7 Tesla by means of a novel multi-variable procedure. A
spectral resolution has been implemented at high field, correlating line-shape
Chapter 5
82
information to relaxation decay. The two-dimensional data sets obtained were analyzed
in terms of two or three decaying components without any assumptions about the
underlying line shape. These results are complemented with classical two and three
component analyses based on the maximum intensity of the FID’s.
5.1. Introduction
Unsaturated polyester (UP) resins are the third-largest class of thermoset molding resins.
They constitute one of the most important matrix resins for glass fibers reinforced
composites. These materials are easy to process and are produced at low production
cost. UP-resins are produced in a two-step process: first, saturated and unsaturated
diacids or anhydrides are reacted with diols via a polycondensation reaction. Secondly,
the resulting polyester is dissolved in styrene monomer. The resin is then processed into
a three-dimensional polymer network (using a peroxide as free-radical initiator) through
a polymerization between styrene and the double bonds in the polyester chains (Figure
5.1). The cross-link density of the network will control the macroscopic properties of
the material. To tailor these properties it is important to look at the molecular structure
and the related molecular mobility of the obtained network.
Mobility in polymer networks has been extensively studied during the last two decades
by proton Nuclear Magnetic Resonance (NMR) spectroscopy, in both solid and swollen
states. NMR has proven to be a valuable tool for studying various types of polymer
networks, including epoxy resins [1], cross-linked polystyrene [2], cured EPDM [3],
cross-linked polyacrylate [4] and unsaturated polyesters [5].
Both NMR as a spectroscopic technique [5,6], and NMR as a relaxometry technique can
be performed on these types of systems; this contribution focuses on the latter
technique. The structure and the dynamics of polymer chains, largely determine the
mechanical properties of polymers. Investigation of these is thus of prime importance
and NMR is an informative non-invasive technique in that regard.
Motions in polymer materials can generally be investigated by NMR relaxometry. NMR
experiments yield information about both the local and long-range mobility of network
polymer chains. One of the most useful NMR techniques for the latter purpose is the
measurement of the transverse magnetization relaxation process, commonly referred to
Chapter 5
83
as T2 (spin-spin) relaxation measurements [7,8,9]. In the solid state, this technique
reflects slow local molecular mobility of polymers. The T2 relaxation times very often
have values smaller than a millisecond.
At such time scale, both chemical cross-links and physical entanglements affect polymer
networks chain’s motion preventing complete motional isotropy. As a result, some
residual dipolar coupling is observed. The proton T2 relaxation experiments are
governed by a de-phasing of coherence influenced by this residual dipolar interaction.
The relaxation rate depends on the strength of the interactions between the nuclear
magnetic dipoles. It has been shown that the magnitude of the residual coupling
increases with the cross-link density of the system [10,11,12,13]. When the chain
motions are less restricted, the interactions will be more averaged out, resulting in a
slower transverse relaxation.
The rate of T2 relaxation can be expressed as the sum of the individual relaxation decays
from different parts of the network without spin-diffusion. To correlate the NMR
relaxation with molecular dynamics, theoretical models are used [14].
The mathematical form of the decay function has been a subject of discussion. A simple
model based on two or more exponential decays is often used as a phenomenological
approach.
Figure 5.1. Representation of the polyester pre-polymer chain cross-linked with styrene.
The main goal of this chapter, presenting results of T2 relaxation NMR experiments on
cured unsaturated polyesters, is to help to better understand the complex chains
dynamics in chemically cross-linked polymer materials. The author tried to probe
Chapter 5
84
heterogeneity of the chain dynamics in terms of two components and three components,
one for describing the rigid part of the system due to the cross-links and the others
describing the more mobile parts.
The experimental data were then first fitted with a simple dual-exponential-decay
approach (below Tg). For a complete and precise description of the system, a spectral
resolution has been implemented, correlating line-shape information to relaxation decay.
The two-dimensional data sets obtained were analyzed in terms of two and three
decaying components without any assumption about the underlying line shape.
5.2. Experimental
5.2.1. Sample preparation
The samples were supplied by DSM Resins (Zwolle, The Netherlands) in the form of a
solution of unsaturated polyester resins (UPR-Palatal intermediate) containing 36.6wt %
styrene. A catalyst (0.5% of a 1% cobalt solution) and 2% of benzoyl peroxide initiator
were added to the solution. The unsaturated polyester is formed by a poly-condensation
reaction of maleic anhydride (MA) and phthalic anhydride (OPA) with 1,2-propylene
glycol (PG) and ethylene glycol (EG). The material was degassed and transferred
between two glass-plates, cured for 24 hours at room temperature, followed by two
post-curing steps for 24 hours each at 60 oC (333 K) and 80 oC (353 K) respectively. For
the NMR experiments, the resulting networks were ground to a fine and apparently
homogeneous powder using a cryogenic grinder (Figure 5.1).
The initial composition of the material leads to the following weight fraction ratio (1;
1.5; 2; 0.5) respectively ( MA; OPA; EG; PG ) and 36% of the total mass being styrene
monomer.
5.2.2. NMR experiments
The NMR experiments were performed on a DMX500 and a MSL200 Bruker
spectrometers operating at 500 MHz and 200 MHz respectively for protons. The
relaxation experiments were carried out at different temperatures on the cross-link
samples between 213 K and 506 K using a variable temperature unit with and without
Chapter 5
85
swelling in Chloroform (CDCl3) (Vs = 40-50%). The low temperature calibration was
performed using methanol. For high temperature calibration ethylene glycol was used
between room temperature (RT) and 390 K. For temperatures around 450 K and higher,
calibration was performed using the melting point of organic compounds. The samples
were equilibrated at each temperature for 20 minutes. The Hahn-Echo Pulse Sequence
(HEPS) was used to record the decay of the transverse magnetization.
A Hahn-Echo (HEPS) measurement is based on a double-pulse sequence: 90°-τ-180°-τ
-acquisition. In our experiments, the τ is varied between 0.01 and 4 ms. The second
pulse in the HEPS inverts nuclear spins and an echo signal is obtained with a maximum
at time equal to (τe) half of the 90° pulse + 2τ + the 180° pulse. By varying the pulse
spacing, intensity profiles are obtained as function of time τe. The 1H-NMR T2 HEPS
relaxation signal is acquired as a function of τe starting from the top of the echo.
5.2.3. Relaxation background
As explained in Chapter 2, mechanical properties of polymer materials are strongly
determined by their chain dynamics.
Chain dynamics in polymer systems, even in a very homogeneous material (one type of
component or phase) is extremely complex and can only be described by a continuous
distribution of correlation times. In polymer networks, as a result of diversity of motions
(Figure 5.2), different parts of the network generate different dynamics, from slow
molecular dynamics (correlation time 10-3 s) to fast molecular dynamics (correlation
time < 10–10 s). Chains dynamics in polymeric network is very often studied by spin-
spin 1H NMR relaxation analysis [1]. This relaxation is induced by the dipole-dipole
interactions between neighbouring spins and is sensitive to anisotropy of long-range
motions. In the liquid state fast isotropic tumbling of the molecules leads to motional
averaging of these dipolar interactions to zero. In the solid state, averaging leads to a
residual part q of the second moment M2 of the dipolar interactions as described in
equation (5.1). This residual coupling is a direct result of the anisotropy of molecular
motions.
Several theoretical models have been developed based on discrete correlation times in
order to fit the transverse relaxation decay. It is accepted that isotropic rotational chain
motions (q=0) exhibit an exponential decay relaxation behavior (BPP theory), whereas
Chapter 5
86
anisotropic chain motions (q≠ 0) are better described by Gaussian decay functions
(Anderson-Weiss theory).
Figure 5.2. Representation of different mobile molecular parts in polymer network: 1) sol fraction, 2) dangling chains ends; 3) inter cross-link chains.
In general, “the relaxation picture” in polymer networks (above Tg) exhibits different
molecular mobilities in the material producing different relaxation signals of different
forms and lengths. This diversity of motions (combination of isotropic and anisotropic
motions) leads to describe the relaxation with a superposition of exponential decays or
stretch exponential decay (Weybull [15]) and Gaussian decays. However, there is still
some discussion in the literature on the mathematical form of this decay.
A more general model that collects with both exponential and Gaussian is widely used
to describe the relaxation of the transverse magnetization in polymer networks [16, 17]:
CBcccA TtCTtBttqMTtAtM 222
22 expexp1expexp)( (5.1)
where M(t) represents the magnetization function, the factor A represents the fraction of
inter-cross-link chains, B corresponds to the fraction of more mobile components such
dangling chains, and C corresponds to the liquid-like fraction commonly named sol-
fraction. t is the pulse spacing in the HEPS experiment. T2A, T2B and T2C are related to
the relaxation decay, τc is the apparent correlation time of inter-cross-link orientations,
Chapter 5
87
and q is a measure of the mean residual part of the second moment M2 of the dipolar
interaction in the rigid limit (T< Tg).
5.2.4 Fitting procedure
The spin-spin 1H NMR relaxation in polymers is usually analysed by recording the
maximum of the Hahn- Echo in the time domain in function of the echo time (t) without
any spectral resolution . Hahn-Echo spectra are recorded by means of a 90°- -180°-
pulse sequence with a 90° and a 180° pulse separated by a variable time interval . The
90° pulse produces initial transverse magnetization, which difocuses during time
under the influenceof spin-spin coupling, chemical shift differences and magnetic
field inhomogeneity. The 180° pulse refocuses the effect of chemical shift and field
inhomogeneity, but leaves the spin-spin coupling unchanged. As a result an echo arises
at time τe which is attenuated only by the spin-spin relaxation. The NMR signal is
acquired as a function of time t starting from the top of the echo:
(5.2) )exp()|()(),( 000inhom0 titRGdtS ee
where Ginhom(0) reflects the inhomogeneous spectrum arising from residual anisotropy,
chemical shift differences and magnetic field inhomogeneity, and R(0| e +t) the spin-
spin relaxation decay associated with spectral frequency 0. In the usual approach, the
top of the Hahn-Echo at t = 0 is measured as a function of e without spectral
resolution. The resulting monotonous decay, a 1D data set, can be analyzed in terms of a
number of components (exponential or Gaussian). We propose to use the spectral
resolution in the full 2D data set I( e ,) after Fourier transformation to the frequency
domain (Figure 5.3, equation (5.3)).
By defining pure components to have mono-exponential relaxation and relaxation-
invariant NMR lineshapes, we impose a mathematical structure of the form:
(5.3) )/exp()(),( 2 ke
n
kke TgI
Chapter 5
88
with gk() the relaxation-invariant subspectrum of the component with relaxation time
T2k.
Figure 5.3. Representation of the data of matrix Y in a two-dimensional fitting approach.
The transition from equation (5.2) to equation (5.3) is not trivial. First, we change from
a realistic continuous description to the discrete picture that can be feasibly extracted
with numerical methods from data with significant noise without some a priori model
for Ginhom(0). Second, by assuming relaxation-invariant lineshapes gk(0) we have
tacitly separated R(0| e +t) into the two time domains according to the Laplace
relation:
)](exp[),(r)|( 00 tdtR ee (5.4)
where r(0, ) denotes the Laplace spectrum of the relaxation decay for the component
with NMR frequency 0. Third, instead of labelling the components on the basis of their
frequency 0, equation (5.3) implicitly classificies spectral components on the basis of
their relaxation rate. To clarify this point we insert equation (5.4) into equation (5.2) and
apply the Fourier transform along t and change the order of the integrals over 0 and :
(5.5) ])(/[2),(r)(]exp[),( 20
200inhom0 GddI ee
Chapter 5
89
The meaning of r(0, ) as the Laplace density function of the component with
frequency 0, has now become the spectral density function of components with
relaxation rate. Equation (5.5) also shows the Lorentzian line broadening resulting
from the relaxation during the acquisition time t. Translated into the discrete picture of
equation. (5.3), equation (5.5) defines the sub-spectrum gk() as:
(5.6) ]2)(1/[2),(r)()( 22
021
200inhom0 kkkk TTTGdg
For a mixture of components with mono-exponential relaxation and well-separated
spectral frequencies and relaxation rates it should be relatively easy to extract the
intrinsic properties from a series of Hahn-Echo spectra. In real cases, however,
components may strongly overlap in both the spectral and Laplace domain. To be able
to distinguish between the different components of the sample a data analysis approach
based on multivariate statistics has to be used. This approach uses a “Multivariate
Analysis” to estimate the line shapes of the overlapping proton resonance. There are
several methods to fit the exponential decay model, which have been used with different
degrees of success, the one used here, is based on the so called LINOL method
describes in more details in Appendix 1 & 2 (Linear-NOn-Linear fitting) [18].
5.2.5. MDSC experiments
It was shown that modulated DSC (MDSC) allows for deconvolution of signals from
amorphous and ordered phases. The method has a greater resolution then non-modulated
DSC. The amorphous phase gives rise to a reversing heat flow, which is readily
converted to specific heat capacity (Cp) and its derivative, which highlights the glass
transition temperatures.
The Tg’s and the corresponding heat capacity of the polyester network were measured
via (MDSC) using TA Q1000 DSC, equipped with an auto-sampler and a refrigerated
cooling system (RCS). The temperature calibration was obtained using Indium. The
sample was prepared in an aluminium pan (between 10 and 15 mg). An oscillating heat
flow signal with a period of 60 seconds and amplitude of 1 oC and an underlying heating
rate of 1 oC/min was used.
Chapter 5
90
5.2.6. DMTA experiments
The glass temperature and the storage modulus were measured with DMTA using a
multi-frequency sweep. Test bar sample of specific dimensions were cut out of the cured
film. Thickness of the cured bar was measured with a calibrated measuring device. The
measurements of storage modulus, loss modulus and tangent delta were carried out at a
frequency of 1 to 100 Hz. Experiments were started at 10 C up to 200 C with a ramp
speed of 5 C /min.
5.2.7. Dielectric experiments (DE)
The dielectric measurements were done using the Novocontrol alpha analyser.
Two gold-plated electrodes with standard configuration were used. The sample
thickness was set about 2 mm, and a layer of gold was applied to the material by
sputtering. The measurements were performed within a broad frequency range (0,125
Hz up to 42 MHz). Experiments were started at 10 C up to 200 C with a ramp speed
of 5 C /min.
5.3. Results and discussion
5.3.1. Network heterogeneity and temperature effect on T2 measurement
5.3.1.1. The experimental overall lineshapes and Hahn-Echo decays (both as 1D
data sets) as a function of temperature
Figures 5.4.a and 5.4.b show the experimental overall line shapes, and Hahn-Echo
decays (both as 1D data-set) as a function of temperature demonstrating the problem of
solving complex polymer dynamics in an unambiguous way. Figure 5.4.a depicts the
influence of temperature on the NMR line shape. Passing the glass transition
temperature (Tg = 80 C), the mobility in the polymer network increases. It is shown
that at higher temperatures (above Tg) more degree of freedom is observed; the motions
are rapid enough and dipole interactions start to average out leading to a narrowing of
the NMR lines. In addition, chemical resolution is improved. For this polymeric system,
Chapter 5
91
the 1H NMR transverse relaxation decay results are presented in Figure 5.4.b. These fits
were obtained with a two exponential decay model, suggesting two types of mobility in
the sample. Fast decays are recorded at low temperature (300 K), while at higher
temperatures (409 K-506 K) slower decays are observed. At higher temperatures (above
Tg) a two exponential decay model seems not to be sufficient anymore.
Figure 5.4.a. (left) NMR line recorded at 2= 16 s ( corresponding to the first point of the exponential decay). Figure 5.4.b. (right) Hahn-echo decays as a function of temperature .
Generally, in chemically cross-link polymeric network, the strong static proton dipolar
interaction makes the spin-spin relaxation process dominant in the relaxation. However,
if the experimental temperature is well above the Tg, the dipole-dipole interactions
become weaker and the relaxation will be dominated by local conformational motions,
narrowing the resonance lines. This will be reflected by much higher relaxation time
values. The extracted values of relaxation times T2 are presented in Tables 5.
It has been already mentioned earlier that the decay of the transverse magnetization, in a
high cross-link density network but also in a rigid matrix (e.g. crystalline phase), is
generally described by a Gaussian function at very short Hahn-Echo times (10 to 30
microseconds), especially at temperatures below the Tg. For the experimental data
below the Tg additional fitting experiments were performed and the data were best fitted
with a three component GGL- model (Figure 5.5).
Chapter 5
92
Table 5.1. T2 relaxation times obtained with 2 exponential decay fit for the cross-linked polyester resins recorded at 500MHz.
T(K) 2K T2s (μs) T2l (μs)
298 9 64
362 12 101
393 26 270
441 51 563
464 84 910
506 109 1083
Table 5.2. T2 relaxation times obtained with three exponential decay fit for the cross-linked polyester resin recorded at 500MHz (2D fit analyses).
T(K) 2K T2s (μs) T2m (μs) T2l (ms)
298 9 64
362 11 65 0.7
393 19 112 0.6
441 33 198 1.8
464 48 284 4
506 62 338 7
Table 5.3. T2 relaxation times obtained with GGL fit for the cross-linked polyester resin recorded at 500MHz (1D fit analyses).
T(K) 2K T2s (μs) T2m (μs) T2l (μs)
298 10 23 155
362 10.7 31.9 140
393 10.2 39 276
It is generally admitted, that for highly cross-linked polymeric networks below the glass
transition temperature (Tg), the broad component should depict a T2s relaxation time in
the order of s. In the present case (Tables 5), the T2s relaxation data show T2 relaxation
times for the broad component varying from 10 to about 100 s as a function of
increased temperature. The increase of the short relaxation time value (T2s) above Tg is
an argument for the absence of crystallinity as well as in favor of network chains
heterogeneity.
Chapter 5
93
Figure 5.5. Three component LLL, GLL and GGL fittings with GGL the most optimal fit. This results for the sol fraction 1) T2l= 276 s with 43,7% fraction estimste for the dangling chains ends 2) T2i=39 s with 27 % estimated for the inter cross-link chains 3) 10.2 μs with 29.4% (at 393 K calibrated temperature).
For all temperatures above the glass transition, the best fit was found using three
component decays following a LLL-best fit (Table 5.2), suggesting that the system at
higher temperature is best described by three types of 1H mobility (Figure 5.2); (1)
protons that are part of the rigid cross-link chain segments which show a fast relaxation
and a T2s in the order of tens of microseconds; (2) protons that are part of network
defects, such as dangling chain ends T2i and (3) protons from sol fractions with groups
that depict a much longer relaxation with a T2l-value of a few hundreds of microseconds
to the ms. This multi-relaxation behavior confirms network heterogeneity. In addition to
NMR, Dynamic Mechanical Thermal Analysis (DMTA), Dielectric Spectroscopy (DES)
and Modulated Differential Scanning Calorimetry (MDSC) experiments have been
performed. In DMTA results, as seen in Figure 5.6, two types of relaxation can be
observed. In terms of mechanical behavior, transitions occur in such polymer materials
when the energy thresholds for certain molecular motions are exceeded. This increased
Chapter 5
94
molecular mobility, caused by the increase in temperature, leads to an overall softening
of the material. This process usually manifests itself as a drop in the storage modulus
and a peak in the loss modulus. In the polyester network, two peaks are observed; one
peak generally assigned to long range chain molecular motions (so called relaxation,
associated with the glass transition), and a secondary relaxation also known as local
mode relaxation, or relaxation, due to small-scale motions (side chain motions and/or
rotations of terminal groups).
Figure 5.6. DMTA loss modulus (multi-frequency sweep) recorded for the cross-linked polyester.
The dielectric spectroscopy (DES) data presented in Figure 5.7 show a very broad Tg
regime between 50 and 100 C. Beyond that temperature range, the dielectric constant
shows an additional relaxation mechanism at the lowest frequencies. Often, this kind of
relaxation mechanism is indicative of certain heterogeneity of the material (presence of
at least two phases with different electrical conductivity).
Chapter 5
95
Additional modulated MDSC experiments have also been performed and two-glass
transitions temperatures were depicted (Tg = 60 C; Tg = 89 C) as seen in Figure 5.8.
All these complementary information and techniques probe and confirm the network
chains heterogeneity.
Figure 5.7. Dielectric measurements. Dielectric constant versus temperature at different frequencies (0.125 Hz, 0.5 Hz, 1 Hz, 256 Hz, 4 KHz, 33 KHz, 2 MHz, 41 MHz; ranging from low frequencies to high frequencies, following the arrow).
In solid-state NMR relaxation experiments, models for the curve shape are still under
debate. In general, physical interpretation of these results is very complex, especially
when fitting two or more relaxation components with weakly separated characteristic
times to a monotonous decay. One way to overcome these ambiguities and to be able to
understand a little further chain dynamics in polyester networks is to combine spectral
and relaxation resolution by correlating line shape information and Hahn-Echo T2H-
relaxation experiments (see Appendix 1 & 2 for more detail).
Chapter 5
96
Figure 5.8. MDSC heating run of the polyester network (top, reversing heat flow vs. T(oC); endo up). Bottom, Reversing Cp vs. T( oC).
Chapter 5
97
5.3.1.2. Combined spectral and relaxation resolution in the 2D-data sets
In Figures 5.9, different experimental Hahn-Echo spectra just above the glass transition
temperature, are represented in function of (half Hahn-Echo time) at respectively 16
μs, 210 μs and 2 ms. An additional line shape is represented and obtained by a
subjective subtraction of the first two Hahn-Echo spectra, obtained at 16 μs and 210 μs.
Some adjustment of the scaling parameter of the two spectra has been done in such a
way that smoothing of the line was obtained. From this two-dimensional approach (2D),
different information could be gained.
First of all, the spectra recorded at 16 μs, 210 μs, and 2 ms show variations in the line
shape, which is already narrowing at 200 μs. This seems to be in agreement with the
relaxation data obtained in Table 5.1, suggesting a T2 relaxation for the broad
component around 100 μs at high temperature. Furthermore, one could observe that on
the spectrum recorded at 2 ms three peaks could easily be assigned. The first two
resonating around respectively 1.3 ppm (-CH3 from PG and CH2- and CH- from PS) and
3.5 ppm (-0CH2 and -0CH from EG and PG) could be assigned to aliphatic protons and
the one at 7 ppm to aromatics. This seems to indicate that the 1H of the polyester
backbone chains as well as the polystyrene chains show the same type of mobility. The
third peak corresponding to the aromatic protons could be attributed both to the
aromatic protons present in the polyester backbone and to the protons present in the
polystyrene cross-link junctions. Also the relative peak intensities of the broad
component at 16 s (at such temperature) closely resemble the envelope of the three-
peak pattern of the mobile component at 2 ms.
This observation could already suggest that the chemical composition of the rigid and
the mobile phase should be relatively close. This can be confirmed by the calculated
proton mol fractions of the aromatic, glycolic and aliphatic protons based on the initial
sample composition (see Appendix 3 of the thesis). The respective calculated values of
33.7, 40 and 26.4 %, correlate quite well with the estimated peak area of the spectrum at
2 ms in Figure 5.9.a recorded at 385 K.
Chapter 5
98
Figure 5.9.a. (left) NMR lines recorded at 2 = 16s, 0.2ms, 2ms (corresponding to the first point of the exponential decay). Figure 5.9.b. (right) Hahn Echo decay recorded at 385 K corresponding to the point selected on the line (at various intensities).
Figure 5.9.b, presents the spectrally resolved T2-behavior. One could see that the
aromatic, glycolic and aliphatic protons depict a similar slower T2 relaxation compared
to the T2 evolution corresponding to the traces and at the left and right side of the
broad signal of the more rigid phase. This observation confirms that the chemical
composition of the mobile and the immobile components in the polymeric system is
closely the same.
5.3.1.3. Multivariate algorithm ("model free", without lineshape assumptions)
Applying the Multivariate Algorithm approach described in the Appendix 1 & 2 of this
thesis, so called "model free" approach, (without line shape assumptions) should help to
probe more objective information.
Figure 5.10.a, shows modeled peaks (at 407 K) obtained for the broad component
SA(W) and the mobile one SB(W). The line shape of the broad component SA(W) seems
to perfectly corresponds to the one in Figure 5.9.a obtained by a subjective subtraction
of the first two Hahn-Echo spectra at respectively 16 μs and 200 μs. Figure 5.10.b shows
the corresponding simulated relaxation decays that assumes that for SB(w), all the
protons (aliphatic, glycolic and aromatics) have the same time decay within the mobile
Chapter 5
99
component. This is in good agreement with what is observed in Figure 5.9.b. The two
approaches (1D and 2D) lead to the same results.
Figure 5.10.a. (left) modelled peaks obtained for the broad and the mobile component obtained at temperature 407K (left). Figure 5.10.b. (right) T2 simulated plots obtained for the broad and the mobile component t(right).
There is an easy way to mathematically check the correspondence between the broad
and narrow line shape by artificially broadening the narrow component. Ideally one
could use the difference in T2 values, because the homogeneous line width is inversely
related to the T2 value. The precise procedure would be to transform the narrow-
component line shape back into the time domain (inverse FT), then multiply with an
exponential decay with decay rate (1/T2s - 1/T2l), and finally transform back into the
frequency domain (FT). Performing such mathematical transformation confirms that the
line shape of the broad component (at such temperature) corresponds to the envelope of
the three-peak pattern of the mobile components.
5.3.1.4. Relaxation in the swollen network
The influence of temperature on T2 relaxation measurements and molecular mobility of
non-swollen cross-linked polyesters resins samples are shown in Tables 5.1, 5.2 & 5.3
and in Figure 5.11.a. The relaxation times of the broad and the narrow components
increase with temperature. It is well known that when temperature increases, the
molecular mobility of network chains is gradually decoupled from that of network
defects. This effect is strengthened at temperatures above the glass transition and results
Chapter 5
100
in a major distinction in the relaxation behavior. The slight increase of the short
component is often attributed to disentanglements of physical cross-links but also to the
increase in frequency of ‘large spatial scale’ motions. The slight decrease in the
strength of inter-chain proton dipole-dipole interactions has also an influence. At some
point, T2s should reach a maximum value (Figure 5.11.a).
Figure 5.11.a. T2 vs. temperature recorded at 500 MHz; correspond to the T2s of the
rigid (short) component, correspond to the T2l of the mobile (long) component.
In general, this is better observed in swollen samples as can be seen in Table 5.4 and
Figure 5.11.b. The physical constraints imposed by the cross-links lead to an
independence of T2 at high temperature (well above Tg) and a plateau is observed (T2pl).
A maximum value of T2s is reached. This value is in the order of magnitude of what is
generally observed for very rigid non-swollen networks around 50 s (Table 5.2).
One could precise that this maximum T2s value is generally used to estimate the cross-
link density of a network by relating the T2pl to the number of statistical segments
between network junctions (equation 5.7). The number of statistical segment between
cross-links can then be used to estimate the average molar mass between cross-links Mc
(equation 5.8) as derived by Gotlib et al. [10]:
Chapter 5
101
Figure 5.11.b. T2s relaxation times for the rigid component at 500 MHz for cross-linked polyester resin swollen in CDCl3.
Z = (T2pl) / [k (T2
rl) ] (5.7 )
Mc = Z c M ru / N (5.8)
where k is the theoretical coefficient, which depends on the angle between segment axis
and the inter-nuclear vector for the nearest nuclear spins. If the spin-spin connecting
vector and the segment direction are perpendicular to each other, then k= 3/10; if they
are parallel k=3/5 [18]. T2rl is admitted to be the T2-value below the glass transition
temperature of the rigid lattice whereas T2pl is the T2- plateau value well above the glass
transition temperature. The maximum of error in the determination of the cross-linked
density by this theory is generally estimated at 20-35% [10]. The biggest assumption is
made in k, and c (representing the number of backbone bonds in one KUHN segment).
It is admitted to be equal to 2 for freely rotating chain and around 10 for polystyrene in
cyclohexane.
Chapter 5
102
In the present unsaturated polyester network, the average molecular mass between
polystyrene (PS) cross-links was estimated for (Mru / N = 52), k= 3/10 and c = 10. The
obtained average molecular mass between PS cross-links (Mc ≂ 482 g/mol) corresponds
to an estimated length of 4.6 styrene monomer units between cross-links.
The mean length between the cross-links in the polyester/PS network can also be
estimated starting from the monomer fractions (see simulation in Appendix 3). From
the calculated proton mol fractions, a mean length of 5.28 styrene monomer units is
estimated for the PS cross-links, supposing a complete reaction of the MA-units. The
estimated mean length between the cross-points in the polyester backbone is 5.79
monomer units. This is rather short and close to the value of 5.28 styrene units for the
PS cross-links. From these short cross-link lengths we do not expect crystallization to
occur. The values obtained through the simulation in Appendix 3 and the one calculated
from Gotlib’s approximations are very close.
In the non-swollen sample, the increase of the long component T2l is much more
pronounced and should keep on increasing with temperature, as it is generally originated
from the relaxation of network defects such as dangling chains end, free chains that are
not attached to the network, chain loops and also to the relaxation of side chain groups
such as -Me groups or (-Ǿ) rings.
Table 5.4 presents T2l from 220 K and up for the swollen sample. A constant value of
about 5 ms is depicted from 300 K to 400 K. Surprisingly, higher values are observed at
low temperature (below 300 K). This effect is generally observed when non-
homogeneous swelling of the network at low temperature occurs or when specific
solvent interactions with the polymer chains are present. A reduction of molecular
mobility is in this case probably a consequence of solvent induced organization [19].
5.3.2 Quantification of rigid and mobile components
Since the integral of the NMR peak is directly proportional to the number of protons in
the material, the relative overall quantities of the network attributed to the rigid
component and the more mobile ones can be calculated in a straightforward way.
Chapter 5
103
Table 5.4. T2 relaxation times obtained with three exponential decay fit for the swollen cross-link polyester resin recorded at 500 MHz.
T(K) 2K T2s (μs) T2i (μs) T2l (ms)
223 11 105 17
253 16 156 12
270 20 192 9
298 26 204 2
320 34 281 5.6
351 42 316 4
385 34 280 5
It is interesting to look at the overall fraction of components in function of temperature.
Tables 5.5 and 5.6 depict the changes in relative fraction of components as a function of
temperature obtained from experiments probed at 500 MHz in the original sample (non
swollen) and in the swollen sample. The main component in the system is undoubtedly
the rigid one representing at room temperature about (80%) of the total system. As
expected from such polymeric system, at high temperature, the amount of rigid material
is decreasing (Table 5.5). Consequently the amount of mobile component in the network
increases. The fraction of component in the non-swollen system seems to be
temperature dependent. This dependence would be surprising if we considered the
system as an ‘ideal’ rigid chemically cross-linked network described by only two
components. This deviation from the ideal network and the temperature dependence on
the fractions of components suggest that the chemical network seems to hold not only
rigid cross-links segments but also physical entanglements.
Also, one can see that the two-component model is too simple to describe this type of
material. It assumes that all the protons have the same time decay and the same NMR
peak shape within the same component (mobile or rigid), which is just not realistic.
With such an approach, small relaxation differences between different protons
belonging to the same component are not considered. This assumption does not
constitute a big drawback at lower temperatures, where the low mobility of the system
yields a broad NMR peak for each component. In this case, the peak shape and the
relaxation decay are dominated by residual anisotropy probed in a sub-millisecond time
and a simple two-component model is enough to describe the system.
Chapter 5
104
Table 5.5. Fractions of components obtained with two (298 K ) and three exponential decay fit for the non-swollen cross-link polyester resin recorded at 500 MHz.
T(K) 2K Rigid fraction
(%) Intermediate fraction
(%) Mobile fraction
(%)
298 91 9
362 78 21 1
393 64 28 8
441 61 33 6
464 57 37 5
506 54 42 4
Table 5.6. Fractions of components obtained with three exponential decay fit for the swollen cross-link polyester resin recorded at 500 MHz.
T(K) 2K Rigid fraction
(%) Intermediate fraction
(%) Mobile fraction
(%)
223 80 12 8
253 72 20 8
298 60 30 10
320 61 33 6
351 59 35 6
385 61 32 7
At higher temperatures however a different panorama is observed. When the
temperature increases well above Tg, the main source of local field seems to be residual
susceptibility and chemical shift rather then residual dipolar field. In that case, mobility
keeps on increasing with temperature resulting in a drastic reduction of the NMR
linewidth. Therefore, the differences among the different protons present within the
same component become more evident, since the peak shape is becoming influenced not
only by the mobility of the system but also by the chemical shift. At 450 K and 506 K,
the intensity of the Me-signal of the propylene glycol (PG) has an increased value at 2=
4 ms, due to its fast conformational rotation, compared to e.g. the aromatic signal of PS
and OPA (shorter T2). This makes the differences in relaxation decay of the different
protons of the same phase more significant. In other words, above a certain temperature
(generally the glass transition temperature), the chemical shift becomes more relevant;
Chapter 5
105
(b)(a)
(Hz)
Inte
nsi
ty (
a.u
.)
(c)
(b)(a)
(Hz)
Inte
nsi
ty (
a.u
.)
(c)
Figure 5.12. Error computed as the difference between experimental and modeled data of T2 experiments recorded at 500 MHz (resp. at 297 K (a), at 393 K (b) and 463 K (c)).
(d) (e)
(f)
(d) (e)
(f)
Figure 5.13. Error computed as the difference between experimental and modeled data of T2 experiments recorded at 200 MHz (resp. at 297 K (d), at 393 K (e) and 463 K (f))
Chapter 5
106
the overlap between the NMR lines is reduced. The failure of a too simple (two-
components) model becomes then more evident. This certainly biases the solution: the
higher the temperature, the higher the error as can be observed in Figures 5.12 and 5.13.
In Figure 5.12 and Figure 5.13, the error computed as the difference between
experimental and modelled data of T2 experiments recorded at 500 MHZ and 200 MHz
is represented. One could see that the lower the field, the smaller is the error, also the
higher the temperature, the more systematic becomes the error.
The differences in fraction of the components could have another origin as well, related
to changes in the composition of the material itself. Effectively, after performing the
measurements up to 500 K, cooled down the sample to 297 K and measured again, a
change in the shape of the relaxation decay was observed and then different T2s and T2l
values were obtained (see Figure 5.14). It seems that the system becomes slightly more
rigid. This may suggests that some changes occurred in the material.
Figure 5.14. Post-curing effect at room temperature (+ correspond to the decay recorded at 500 MHz at room temperature prior to temperature treatment. correspond to the decay recorded at room temperature after temperature treatment).
Chapter 5
107
This could be attributed to different reasons such as, ageing of the system, post-curing
of the system, or, rearrangements occurring in term of phase separation. Effectively,
above the Tg the mobility of the system increases and the system can vitrify some more
on that same time scale.
5.4. Conclusion
The 1H spin-spin NMR relaxation data show a strong heterogeneous dynamic behavior.
The data obtained via other characterization techniques such as DMTA, DES and
MDSC confirmed this result. A combination of gaussian and exponential decays were
required to fit the experimental data. A simple dual exponential decay fit generally used
in affine homogeneous networks with low cross-link density was just not realistic.
Three types of 1H mobility were depicted above Tg, and combining high spectral
resolution NMR experiments to 1H-NMR relaxation experiments (2D data analysis)
helped for a better understanding of the network chain dynamics.
Finally, the average molecular weight between cross-links was estimated and found to
be relatively low (only a few polystyrene monomer units). However, the use of models
to estimate the average molecular weight between cross-links implies to take the
quantitative aspect of that work with some care.
Chapter 5
108
5.5. References
[1] Litvinov, V. M.; De Prajna, P. Spectroscopy of Rubbery Materials, RAPRA Technology 2002.
[2] O’Donnell, J. H.; Whittaker, A. K. Polymer 1992, 33, 62.
[3] Mohanraj, S.; Ford, W. T. Macromolecules 1985, 18, 351.
[4] Winters, R.; Lugtenburg, J.; Litvinov, V. M. Polymer 2001, 42, 24.
[5] Spyros, A. Journal of Applied Polymer Science 2003, 88, 1881.
[6] Graf, R.; Demco, D. E.; Hafne, S. H.; Spiess, W. Solid State Nuclear Magnetic Resonance 1998, 12, 139.
[7] Heuert, U.; Knorgen, M.; Menge, H.; Scheler, G.; Schneider, H. Polymer Bulletin 1996, 37, 489.
[8] Kuhn, W.; Barth, P.; Denner, P.; Muller, R. Solid State Nuclear Magnetic Resonance 1996, 6, 295.
[9] Fisher, E.; Grindberg, F.; Kimmich, R. Journal of Chem. Phys. 1998, 109, 846.
[10] Gotlib, Y.; Lifshitz, M. I.; Shevelen, V. A.; Lishanskij, I. S.; Balanina, I. V. Polym. Sci. 1976, 491.
[11] Simon, G.; Schneider, H.; Hausler, K. G. Prog. Colloid Polym. Sci. 1988, 78, 30.
[12] Simon, G.; Schneider, H. Polym. Bull. 1989, 21, 475.
[13] Fedotov, V. D.; Tshernov, V. M.; Wolfson, S. I. Polym. Sci. 1978, 260.
[14] Brereton, M. G. Macromolecules 1990, 23, 1119.
[15] Bloembergen, E. M.; Purcell, R.; Pound, V. Physical Review 1948, 73, 679.
[16] Litvinov V. M.; Dias, A. A. Macromolecules 1999, 32, 3624.
[17] Simon, G.; Baumann, K.; Gronski, W. Macromolecules 1992, 25, 3624.
[18] Ruytinx, B.; Berghmans, H.; Adriaensens, P.; Gelan, J. Macromolecules 2001, 3, 552.
[19] Menge, M.; Hotopf, S.; Heuert, U.; Schneider, H. Polymer 2000, 41, 3019.
[20] Orza, R. A.; Magusin, P. C. M. M.; Litvinov, V. M.; van Duin, M.; Michels, M. A. J. Macro. Symp. 2007, 40, 899.
Chapter 6
partly reproduced from: Ziari, M.; Asselen Van, O. L. J.; Jansen, M. A. G.; Goossens, J. G. P.; Schoenmakers P. J. Macromolecular Symposia 2008, 265, 290.
109
Chapter 6
A FTIR study on the solid-state copolymerization of bis(2-
hydroxyethyl)terephthalate and poly(butylene terephthalate) and the
resulting copolymers
Summary
The aim of this work was to study the solid-state copolymerization (SSP) of bis(2-
hydroxyethyl)terephthalate (BHET) with poly(butylene terephthalate) (PBT) by FTIR
spectroscopy. The development of the chemical microstructure during the SSP-reaction
was examined as a function of the BHET content, showing the different regimes.
Depending on the ratio of PBT to BHET, a competition exists between annealing of
PBT, incorporation of BHET in amorphous PBT and BHET homopolymerization.
The thermal behaviour of the resulting copolymers was also investigated using infra-red
dynamic spectra. For low BHET-concentrations, only crystallization of PBT-sequences
was observed, while for high BHET-concentrations, only crystallization of PET-
sequences was detectable with a cross-over behaviour for intermediate concentrations.
Chapter 6
110
6.1. Introduction
Poly(butylene terephthalate) (PBT) and poly(ethylene terephthalate) (PET) are two
semi-crystalline polymers that are used in many engineering applications. The main
advantage of PBT is its high crystallization rate, making it suitable for injection molding
applications with short molding cycle times [1]. PET has a considerably lower
crystallization rate and is therefore mainly used for fiber applications and packaging [2].
The glass transition temperature of PBT is rather low compared to PET (Tg 45 °C for
PBT and Tg 80 °C for PET). To obtain a Tg higher than 45 °C, PBT and PET can be
reactively blended in the melt. The concomitant ester-interchange reactions occurring in
the melt first result in the formation of block copolymers, but as the reaction proceeds,
random PBT-PET copolymers are obtained [3-5]. These random PBT-PET copolymers
have a Tg in between that of the PBT and PET homopolymers. However, the shorter and
more irregular homopolymer sequences of these random copolymers consequently lead
to a lower melting temperature, crystallization rate and crystallinity with respect to pure
PBT. It is therefore desirable to synthesize PBT-PET copolymers having a Tg higher
than the PBT homopolymer, but with a crystallization behavior comparable to that of
PBT. Therefore, a copolymerization method should be used that enables to retain large
crystallizable homopolymer PBT blocks with a non-random chemical microstructure.
In previous work [6], PBT-PET copolymers were made by copolymerization in the melt
(MP) and the solid state (SSP) of bis(2-hydroxyethyl)terephthalate (BHET) with PBT.
The chemical microstructure of the synthesized PBT-PET copolymers was examined by 13C-NMR spectroscopy sequence distribution analysis. As expected, the chemical
microstructure of the PBT-PET copolymers obtained by MP was fully random.
However, when BHET was incorporated into PBT via SSP, a non-random blocky
chemical microstructure was obtained. When the fraction of BHET used for
incorporation in PBT was increased, the blocky character of the chemical microstructure
became more pronounced, suggesting that BHET may react by self-condensation to
form homopolymer PET blocks [7, 8]. The 13C-NMR sequence distribution analysis also
showed that transesterification reactions occurred between BHET monomer and PBT. It
was not clear whether the formed homopolymer PET blocks are present as a separate
phase or whether these blocks reside in the amorphous phase of PBT.
Chapter 6
111
In this chapter, the incorporation of BHET into PBT via SSP is studied in more detail.
The kinetics of the incorporation via SSP is studied under isothermal conditions by
using FTIR spectroscopy. During the SSP process, a competition between annealing and
dissolution of the PBT-crystals, incorporation of BHET in the amorphous PBT and
BHET homopolymerization exists, of which the relative rates depend on the ratio of
PBT/BHET. In a previous study, the miscibility of the BHET with PBT and the thermal
properties of the resultant BDxEGy-copolymers obtained via SSP were investigated by
using DSC [9]. It was observed that BHET is partially miscible in the amorphous PBT,
depending on the PBT/BHET ratio. The crystallization behaviour of the resultant
BDxEGy copolymers, where BDx represents the initial mol% PBT and EGy the mol% of
hydroxyl end groups per mole BHET, was studied by DSC. It was observed that when
more BHET is incorporated the onset of crystallization shifts to lower temperatures,
while the crystallization exotherm becomes broader. However, the (BD30EG70)ssp
copolymer displayed a much sharper crystallization exotherm with a slightly higher
onset. A similar observation was made by Misra et al. [10]. However, DSC lacks the
possibility to investigate the crystallization behaviour of the individual components
when the crystallization exotherms display a large overlap. In this respect, temperature-
dependent FTIR spectroscopy is much more discriminating by following absorption
bands, which can either be assigned to the crystalline or amorphous phase of PBT or
PET.
The SSP kinetics were investigated under isothermal conditions using two
compositions. In that way, the development of the chemical microstructure during the
SSP reaction could be examined together with the miscibility of BHET and PBT.
Temperature-dependent experiments were performed on three different BDxEGy
copolymers and compared to PBT using the dynamic infra-red spectra obtained through
the 2D-IR correlation algorithm developed by Noda et al. [11, 12].
6.2. Experimental
The BDEG copolymers were obtained by incorporation of BHET into PBT by using
SSP. At first, the kinetics of the SSP reactions was investigated with attenuated total
reflection (ATR)-FTIR spectroscopy. A BioRad FTS3000 spectrometer equipped with a
Chapter 6
112
MCT detector was used to record the spectra. The samples were pressed onto the
diamond crystal of a Speak Golden Gate ATR accessory. Next, the sample was heated
to 120 C and the initial spectrum was recorded. Upon starting the experiment, the
sample was heated to 180 C using a heating rate of 30 C/min. Spectra were recorded
during the reaction with a time interval of 60 seconds using a spectral resolution of 4
cm-1, co-adding 62 scans. At 180 C, the sample consists of a semi-crystalline fraction
of PBT and a liquid fraction of BHET. During the initial stage of the SSP-reaction, the
fraction of the liquid BHET in contact with the crystal increases, while the fraction of
PBT decreases. It becomes then very difficult to monitor the reaction quantitatively.
Therefore, it was decided to perform all experiments in the transmission mode. Then, a
more or less constant cross-section of the sample is monitored. These experiments were
performed using a BioRad UMA500 IR microscope coupled to a BioRad FTS6000
spectrometer. A Linkam TMS600 hot stage was used for temperature control. To obtain
a thin sample, the powder samples were cold pressed by a diamond anvil cell. Next, the
sample was placed in between two zinc selenide (ZnSe)-windows and transferred to the
Linkam hot stage. Spectra were recorded with a resolution of 2 cm-1 co-adding 250
scans. The crystallization behaviour of three BDEG copolymers were also followed by
using the same set-up as described above with a cooling rate of 10 °C/min. The samples,
however, were not covered with a second ZnSe window. For these experiments, spectra
were recorded using a resolution of 2 cm-1 co-adding 100 scans. The 2D-Pocha
software, developed at the Kwansei Gakuin University (Japan), was used for the data
treatment. The spectrum at 180 °C was taken as the mean spectrum for the dynamic
spectra calculations.
6.3. Theory
Two-dimensional infra-red (2D-IR) spectroscopy is an analytical technique based on
time-resolved detection of IR absorption to look at molecular interactions. In 2D-IR, a
spectrum is obtained as a function of two independent wave numbers. The general
experimental approach used in 2D correlation is based on the detection of dynamic
variations of spectroscopic signals induced by an external perturbation.
Chapter 6
113
The effect of such induced-perturbations leads to changes of the local molecular
environment and will be manifested by time-dependent fluctuations of various spectra,
yielding so-called dynamic-spectra. Mathematical manipulation of these spectra will
give a useful set of 2D-IR spectra.
There are different types of external perturbations that could be used; the type of
information that can be obtained from the dynamic spectra will depend on the type of
perturbation. A more detailed information on the formalism can be found in NODA
feature article [11]. This approach is used to lift many analytical ambiguities. For
instance, it helps to simplify and understand complex spectra containing many
overlapped peaks. It can also be used to enhance the spectral resolution by spreading the
absorption peaks into a second dimension as much as it could help to identify different
intramolecular interactions through selective correlation of IR absorption bands. In this
chapter, thermal excitation has been used. Temperature-dependent experiments were
performed on three different BDxEGy copolymers and compared to PBT. Hence, the
microstructures of the copolymers could be investigated.
6.4.. Results and discussion
6.4.1. Assignment of the absorption bands of the pure components
Table 6.1 shows the assignment of most relevant absorption bands of PBT, PET and
BHET. For crystalline PBT, a band at 1454 cm-1 (CH2 bending) is observed, that for
amorphous PBT shifts to 1468 cm-1 together with absorption bands at 1320 cm-1 (ring
ester in plane) and 1389 cm-1 (CH2 wagging). Crystalline PET shows characteristic
absorption bands at 1471 cm-1 (CH2 bending), at 1342 cm-1 (CH2 wagging) and at 990
cm-1, of which the latter band is only observed if chain folding occurs. Typical bands of
amorphous PET can be observed at 1371 cm-1 (CH2 wagging), 1458 cm-1 (CH2 bending).
For BHET, a lot of absorption bands overlap with PBT or PET. To follow the
copolymerization, the absorption band at 3550 cm-1 (OH stretching) can be used.
Chapter 6
114
Table 6.1. Assignment of the absorption bands of the pure components [13].
PBT ( cm-1)
PET ( cm-1)
BHET ( cm-1)
amorphous crystalline amorphous crystalline
OH stretching 3550
CH2 bending 1468 1458/1454 1458/1440 1471 1453
Ring-in-plane-def 1408 1409 1409 1409 1407
Ring CH in-plane def 1380
CH2 wagging 1389 1389 1371 1342 1372
Ring CCH 1320 1315
Chain folding 990
6.4.2. Kinetics of the SSP reaction studied by FTIR spectroscopy
Since the spectra of PBT, BHET and PET show overlapping absorption bands, it was
decided to focus on the four following absorption bands: the crystalline PET absorption
band at 1342 cm-1, the absorption band at 1371 cm-1 of amorphous PET or BHET, the
crystalline PBT band at 1454 cm-1 and the OH-stretch vibration band of BHET at 3550
cm-1 at starting time (t = 0)..
The SSP-kinetics at 180 C was followed for the (BD70EG30)feed and (BD30EG70)feed
systems. Figures 6.1a and 6.1b show the individual spectra before and after the SSP-
reactions, while the kinetic results are shown in Figures 6.2a and 6.2b, respectively. The
disappearance of the absorption band at 3550 cm-1 shows that in both cases the BHET is
converted. For the (BD70EG30)feed system (Figure 6.2a), the intensity decreases gradually
and the behaviour is typical for a series of two irreversible consecutive first-order
reactions. This was also observed for the incorporation of 2,2-bis[4-(2-
hydroxyethoxy)phenyl]-propane (Dianol 220) [14]. The (BD30EG70)feed system (Figure
6.2b) shows a different behaviour. After a similar gradual decrease of the intensity up to
approx. 20 min, the reaction accelerates and finally reaches a plateau level after approx.
50 min. In both figures, the absorption band of the PET CH2 wagging mode in the
crystalline state is also depicted, although the band position changes from at 1340 to
1342 cm-1 on going from (BD70EG30)feed to (BD30EG70)feed. Cole et al. studied the
microstructure of PET homopolymer using reflection and transmission FTIR
Chapter 6
(a)(a)
(b)(b)
Figure 6.1. Spectra before (-) and after (..) the SSP-reaction: (a) (BD70EG30)feed and (b) (BD30EG70)feed.
115
Chapter 6
116
spectroscopy and showed that the ethylene glycol segments for PET in the crystalline
phase are always in the mutual trans orientation (T), while in the amorphous phase a
mutual gauche orientation predominantly exists (G) [15]. They also observed that
different conformation states can also exist for the C-O bond of the glycol group, which
can likewise exist in a trans (t) or gauche (g) conformation as well as the terephthalic
acid segment. If the latter is in a non-trans conformation and the ethylene glycol
segment is in a trans (T) conformation, a “rigid” amorphous band will be observed at
1340-1338 cm-1. X-ray diffraction measurements showed that all three groups adopt the
trans conformation (T, t, Tb) in the crystalline phase. Then, an absorption band at 1342
cm-1 of the glycol C-C unit will appear. The change in peak position is a strong
indication that for the (BD70EG30)feed system the BHET is incorporated in the PBT
chain, while for the (BD30EG70)feed system separate PET crystals can be formed. The
difference in behaviour is also found for the 1371 cm-1 band. For the (BD30EG70)feed
system, the same intensity profile is found as for the OH-stretching vibration band of
BHET (3550 cm-1), but (BD70EG30)feed system remains at the same absorption level for
the first 30 min and then starts to decrease up to approx. 40 min followed by a plateau.
This might indicate that during the first 30 min, the decrease of the intensity due to the
conversion of BHET is compensated by the formation of amorphous PET, which also
has an absorption band at that wavenumber.
For the (BD30EG70)feed system, the crystalline PBT band at 1454 cm-1 decreases,
probably because of the decrease of the BHET that also has an absorption band at 1454
cm-1, although this decrease may also be attributed to partial dissolution of the PBT-
crystals that may occur because of the high mole fraction of BHET acting as a solvent
for PBT. On the contrary, for the (BD70EG30)feed system, the band at 1454 cm-1
increases. This behaviour can only be explained by the occurrence of annealing and
perfectioning of the PBT-crystals due to the high mobility during the SSP-reaction.
Chapter 6
117
(a)(a)
(b)(b)
Figure 6.2. Relative intensities of the indicated absorption bands of crystalline PET, amorphous PET, PBT and BHET during SSP of: (a) (BD70EG30)feed and (b) (BD30EG70)feed.
Chapter 6
118
6.4.3. Temperature dependent experiments
Changes in the FTIR-spectra were monitored as a function of temperature. Figure 6.3
shows the dynamic spectra of BDEG copolymers with different PBT/BHET ratios
cooled from the melt in a temperature range from 240 to 170 C. For the PBT
homopolymer, positive bands at 1454, 1386 and 1320 cm-1 develop during cooling.
These bands can be assigned to crystalline PBT (CH2 bending and wagging modes).
Also the positive band at 1408 cm-1 can be assigned to amorphous PBT (See Table 1).
In the spectra of the (BD70EG30)ssp system, beside the positive bands of crystalline PBT,
also a positive band is observed at 1338 cm-1, which can be assigned to CH2 wagging of
an ethylene segment (trans conformation) with the terephthalate segment in a non-trans
configuration. This is an indication that ethylene segments are incorporated into the
PBT chain.
In the spectra corresponding to sample (BD50EG50)ssp the positive bands of PBT are
weaker and positive bands at 1471 and 1342 cm-1 corresponding to crystalline PET
appear. Negative bands can be observed at 1315 and 1371 cm-1, which can be assigned
to amorphous PET, as well as a band at 1465 cm-1, which can be assigned to amorphous
PBT. These spectral features clearly indicate that PET crystals are formed. Furthermore,
the crystalline PBT and PET bands appear almost at the same time. The limited
miscibility of BHET with PBT promotes the formation of separate PET-crystals.
However, no visible absorption band at 990 cm-1 related to PET chain folding is found,
suggesting that the crystalline PET chain segments are not long enough.
In the spectra corresponding to the (BD30EG70)ssp system, weak spectral features of
crystalline PBT are present together with strong positive and negative features of PET.
If we have a closer look at the bands at 1320 and 1454 cm-1 and the band at 1342 cm-1, it
can be noticed that PET crystallizes much earlier, indicating that PBT and PET
crystallize independently and are mostly like in a separate phase. This is confirmed by
the presence of an absorption band at 990 cm-1, that can be assigned to chain folding of
PET.
These observations are in line with the previously reported DSC-results. When more
BHET is incorporated, the onset of crystallization shifts to lower temperatures, while the
crystallization exotherm becomes broader, mainly because the PBT crystallization is
Chapter 6
119
disturbed by the presence of short PET sequences in PBT. However, the exotherm of the
(BD30EG70)ssp copolymer, which is much sharper with a slightly higher onset, is due to
crystallization of PET, while at lower temperatures only a small PBT crystalline fraction
is formed. Most likely, because of the limited miscibility of BHET in the amorphous
phase of PBT and the concomitant formation of PET homopolymer by self-
condensation, the large interfacial area leads to a strong nucleation effect.
Figure 6.3. Dynamic spectra as a function of temperature: 210-170 °C with 10 °C steps: (a) PBT, b) BD70EG30, (c) BD50EG50, and (d) BD30EG70.
Chapter 6
120
Figure 6.4 shows the crystalline PBT absorption bands at 1454 cm-1, and the crystalline
PET absorption bands at (1342 cm-1) as function of temperature after cooling the
samples from 240 C to 110 C. One could observe that crystalline PBT bands decrease
with decreasing fraction of PBT in the (BDxEGy) mixtures in accordance to earlier
results.
In addition, both samples (BD30EG70)ssp and (BD50EG50)ssp show a strong intensity for
the PET crystalline absorption bands at 1342 cm-1.
-1
0
1
2
3
4
100 120 140 160 180 200 220 240
Temperature (oC)
Abs
orba
nce
(a.u
.)
PBT, 1458 cm-1
80% PBT, 1454 cm-1
80% PBT, 1339 cm-1
70%PBT, 1454 cm-1
70% PBT, 1339 cm-1
50% PBT, 1454 cm-1
50% PBT, 1342 cm-1
30%PBT, 1458 cm-1
30% PBT, 1342 cm-1
Figure 6.4. Absorbance of crystalline PBT absorption bands at 1454 cm-1 and crystalline PET absorption bands at 1342 cm-1 as function of temperature (cooling from 240 C to 110 C).
From the intensity of the crystalline PBT absorption bands at 1454 cm-1 and after
normalization (using the band at 1410 cm-1 present in PBT and PET crystals), the
crystallinity of PBT in each sample was estimated. These calculations were based on the
crystallinity of pure PBT (54%). The results are reported below in Table 6.2.
Chapter 6
121
Table 6.2. Estimated values (%) of crystallinity in the sample based on pure PBT crystallinity
Samples (BDxEGy)ssp Crystallinity
(%)
Pure PBT 54
(BD80EG20) 35.5
(BD70EG30) 37
(BD50EG50) 20
The crystallinity values are in good agreement with the earlier study by Jansen et al.
using DSC,for the samples with high amount of PBT, where no separate PET crystals
are expected. For the sample (BD50EG50) a big discrepancy is observed with the DSC
results, and shows a much lower crystallinity. One can notice that the crystallinity
determined via DSC is based on the amount of heat needed to break the totality of
crystals order, whereas the crystallinity determined with FTIR spectroscopy is based on
the intensity of the crystalline PBT absorption band. The much lower value obtained via
IR , for the sample (BD50EG50), may be a net indication of the presence of additional
PET crystals in this sample that are not taken into account in this estimation.
6.5. Conclusions
FTIR spectroscopy in transmission can be used to follow the kinetics of BHET-
incorporation in PBT during SSP and to study the microstructure of the resulting BDEG
copolymer. These results show that depending on the ratio of PBT/BHET, a competition
exists between annealing of PBT, incorporation of BHET in amorphous PBT and BHET
homopolymerization. For the (BD70EG30)feed system, BHET is incorporated in PBT
forming a non-random copolymer, while for high BHET concentrations a separate PET
phase is found.
The dynamic spectra measured during cooling from the melt clearly showed that for low
and intermediate BHET-concentrations, crystallization of PBT is responsible for the
crystallization exotherm, as observed by DSC, while for high BHET-concentrations the
DSC crystallization exotherm is caused by crystallization of PET.
Chapter 6
122
Furthermore, the crystallinity of PBT in each sample was estimated. For samples
containing high concentration of PBT, the results for the crystallinity were found to be
in good agreement with the values measured by DSC. Nevertheless, for the sample with
lower amount of PBT, (BD50EG50)ssp, a significant discrepancy among the results was
found between the two techniques. The much lower crystallinity value obtained via
FTIR for this sample indicates that there are additional crystals of PET. Indeed, the
crystallinity determined with FTIR spectroscopy takes only into account the crystalline
PBT absorption band.
Chapter 6
123
6.6. References
[1] Radusch, H. J.; Handbook of Thermoplastic Polyesters, S. Fakirov Ed. 2002.
[2] Gupta, V. B.; Bashir, Z. Handbook of Thermoplastic Polyesters, S. Fakirov, Ed. 2002.
[3] Backson, S. C. E.; Kenwright, A. M.; Richards, R. W. Polymer 1995, 36, 1991.
[4] Jacques, B.; Devaux, J.; Legras, R.; Nield, E. J. Polym. Sci., Polym. Chem. 1996, 34, 1189.
[5] Kim, J. H.; Lyoo, W. S.; Ha, W. S. J. Appl. Polym. Sci. 2001, 82, 159.
[6] Jansen, M. A. G.; Goossens, J. G. P.; de Wit, G.; Bailly, C.; Koning, C. E. Anal. Chem. Acta 2006, 557, 19.
[7] Tomita, K. Polymer 1973, 14, 50.
[8] Lin, C. C.; Baliga, S. J. Appl. Polym. Sci. 1986, 31, 2483.
[9] Jansen, M. A. G.; Wu, L. H.; Goossens, J. G. P.; de Wit, G.; Bailly, C.; Koning, C. E. J. Pol. Sci., Pol. Chem. 2007, 45, 882.
[10] Misra, A.; Garg, S. N. J. Pol. Sci., Pol. Phys. 1986, 24, 983.
[11] Noda, I. Applied Spectroscopy 1990, 4, 44.
[12] Harrington, P.; Urbas, A.; Tandler, P. J. Chemometrics and Intelligent Laboratory Systems 2000, 50, 149.
[13] Cole, K. C.; Ajji, A.; Pellerin, E. Macromolecules 2002, 35, 770.
[14] Jansen, M. A. G.; Goossens, J. G. P.; de Wit, G.; Bailly, C.; Koning, C. E. Macromolecules 2005, 38, 2659.
[15] Cole, K. C.; Ajji, A.; Pellerin, E. Macromol. Symp. 2002, 184, 1.
124
Appendix 1
partly reproduced from: Vivo-Tryols, G.; Ziari, M.; Magusin, P. C. M. M.; Schoenmakers P. J. Analytica Chimica Acta 2009, 37, 641.
125
Appendix 1
Effect of initial estimates and constrains selection in Multivariate Curve
Resolution – Alternating Least Squares. Application to low-resolution NMR data
Summary
A comprehensive study of the applicability of multivariate curve resolution (MCR) methods
to series of T2-relaxation filtered 1H NMR spectra of a crosslinked polymer network is
presented. A collection of Hahn-Echo NMR spectra is obtained at different echo times,
yielding two-way data. In this study the applicability of two different types of orthogonal
projection approach (OPA1 and OPA2) (column-wise and row-wise) were tested. Four
different strategies of alternating least squares methods were also examined (ALS1, ALS2,
ALS3 and ALS4). These strategies differed on the order of measurement for which the
constraints were applied in the final output, and the way in which SSR was calculated to
monitor for convergence. In the spectral order of measurement, a non-negativity constraint
was imposed, whereas in the time order of measurement, the signal was forced to follow an
exponential decay. This yielded up to eight MCR configurations, giving different results.
For solid-state NMR, the dissimilarity in NMR profiles is significantly lower than the
dissimilarity in signal decays, and therefore OPA2 performed better. A final output with a
constrained solution in relaxation-time was preferred (instead of a constrained solution in
NMR spectra) for practical purposes. Differences between the solutions given from the two
ALS configurations can be interpreted as a sign of lack of fit.
Appendix 1
126
1. Introduction
Multivariate analysis is a modern approach to interpret complex experimental data. The
underlying idea is that a coupled analysis of several data sets together (i.e., the information
coming from multiple channels at the same time) yields significant advantages over
analysing every data set separately. This has been demonstrated in many cases. The more
complex the instruments, the greater the advantages of a multivariate approach.
One factor that affects the selection of the most appropriate method(s) is the order of the
experimental data sets. According to the terminology introduced by Kovalski [1],
instruments can be classified according to the order of the tensor that represents the data
obtained from a single experiment. Zeroth-order instruments yield a scalar (i.e. a zeroth-
order tensor), first-order instruments yield a vector (first-order tensor), second-order
instruments yield a matrix (second-order tensor), etc. Among second-order instruments, a
distinction must be made between bilinear and non-bilinear data. This determines the type
of algorithms that can be used for data treatment. If the data is bilinear, the D matrix
representing the data of the second-order instrument can be decomposed as follows:
εCSD (1)
where C is a matrix having in each column the signal of one compound in the first order of
measurement and S is a matrix with in each row the signal of one compound in the second
order of measurement and represents the noise of the data. Chemically speaking, Eq. (1)
means that the number of sources of variance equals the number of chemical components in
the studied system [2]. For example, for a mixture of r compounds in high-pressure liquid
chromatography (HPLC) with ultraviolet (UV) detection, D would be the two-dimensional
spectrochromatogram of UV spectra (horizontal) as a function of elution time (vertical). The
matrix C contains the r elution profiles (column-wise) and the matrix S contains the r UV
spectra (row-wise) of the r pure compounds. In practice, the pure-component features may
not be known in advance. By means of multivariate analysis, the signals of the individual
(overlapping) compounds can be separated. Even in cases of severe co-elution, the peaks of
each compound can be quantified, provided that the signals (spectra) of the compounds
obtained by the (multichannel) detector are sufficiently different.
Appendix 1
127
Basically, the aim of multivariate-analysis algorithms applied to second-order instruments
yielding bilinear data is to solve Eq. (1), removing the noise of the signal and obtaining
both C and S. These two matrices contain all the individual signals of each pure compound
in the two orders of measurement. Eq. (1) can be solved by several methods, including
Alternating Least Squares [3] or Heuristic Evolving Latent Projections (HELP) [4].
Alternatively, if three-way data is available (which implies access to several experiments
performed in a second-order instrument) Generalised Rank-Annihilation Method (GRAM)
[5] or PArallel Rank-FACtor analysis (PARAFAC) [6] constitute an alternative that
indirectly solves Eq. (1). The common denominator in these techniques is to detect the
different sources of variance in the raw-data matrix (D) and to assign each one to a specific
compound. From among this family of techniques, ALS has become popular. This is
probably due to its simplicity –in programming terms– and to the fact that only one sample
is needed. The latter aspect helps to avoid the need for data alignment. Data alignment
strategies are normally a must when multivariate techniques are applied to the
chromatographic field. This requirement normally arises when more than one
chromatographic separation is simultaneously subjected to multivariate analysis. In ALS, if
more than one sample is available, the data from different experiments can be adjoined
forming an enriched two-way data matrix, avoiding the need for peak alignment.
Two problems must be solved when applying the ALS method to second-order instruments.
First, ALS needs to start with an initial guess for either the C or the S matrix. This is
normally tackled by performing several peak-purity assays on the original data matrix, for
example using Orthogonal-Projection Approach (OPA) [7] or SIMPLe-to-use Interactive
Self-modelling Mixture Analysis (SIMPLISMA) [8]. OPA constitutes an excellent
approach, yielding reasonably good initial guesses for C or S matrices. A second problem to
overcome is the so-called rotational ambiguity [9]. This refers to the fact that multiple
matrix pairs C, S mathematically fulfil Eq. (1), while only one of these represents the true
physical solution. In ALS, this problem is partially solved by imposing constraints that
modify the C and S solutions so as to yield profiles that are chemically meaningful in both
orders of measurement. However, depending on the problem, some rotational ambiguity
may remain even with imposed constraints.
OPA-ALS has been applied successfully to second-order instruments involving a separation
with a multi-channel detector such as HPLC-DAD [10] or CE-DAD [11]. In these cases,
Appendix 1
128
OPA was applied to provide an initial estimate of the spectrum of each compound (S
matrix). Instead, OPA may also be applied to obtain initial estimates of the individual
chromatographic peak profiles (C matrix). In HPLC-DAD the former is preferred, because
the selectivity is higher in the chromatographic domain than in the (UV) spectral domain.
The applications of ALS go far beyond hyphenated chromatographic techniques. The
method has been used, for example, for near-infrared-spectroscopy (NIR) data [12], Raman
spectroscopic images [13], kinetic-reaction data obtained with NMR [14], or Diffusion-
ordered spectroscopy (DOSY-NMR) data [15]. In each application, the constraints applied
to C and S profiles are different, based on (chemical) knowledge in each field.
In this work we apply OPA-ALS to a series of Hahn-Echo 1H-NMR spectra as a function of
the echo time for a cross-linked unsaturated polyester resin. Series of Hahn-Echo spectra are
obtained at various echo times. This yields a second-order data set, in which one of the
orders of the measurement is the conventional NMR spectrum (frequency or chemical-shift
axis) and the other the decay as a function of the Hahn-Echo time. The same kind of data is
obtained in DOSY NMR [15]. The entire signal is often analyzed in terms of a few
components, assigned to different phases or fractions in the sample polymers [16-18]. Each
"component" yields a signal of different shape in the NMR spectrum (first order of
measurement) and signal-decay curve (second order of measurement). The signal decay for
each component is assumed to be exponential. There are various reasons to treat the
outcome of such analyses with care, especially when it comes to the physical interpretation
of the results. First, according to general relaxation theory, exponential relaxation is only
expected for rotational chain motions, which are isotropic and fast at a miliseconds
timescale. Chain motions in rubbery polymers are anisotropic and therefore expected to
produce non-exponential decays [19, 20]. Models for describing the curve shape are still
under debate [21]. Second, chain dynamics are highly complex – even within a single
polymer phase or fraction – and they may be more-realistically described with a continuous
distribution of correlation times. In practice, decay curves are often fitted as a sum of
exponential relaxation curves (mono-, bi- and tri-exponential relaxation decays). The
adequate precision obtained with such empirical models may result from the combined
effects of motional anisotropy and physical heterogeneity.
There are important differences with previously published applications of OPA-ALS that
render the present case more difficult. Firstly, if we try to distinguish between fractions of a
Appendix 1
129
polymer that are chemically similar, but physically different (e.g. crystalline and amorphous
regions in a polymer), then we may expect that the NMR spectra are highly correlated. The
peaks should be located at the same frequencies and only differ in their widths. This calls
for caution: some steps in the ALS algorithm may yield highly unstable results, since the S
matrix (containing NMR spectra) will be nearly singular. As it has been discussed in other
areas (like imaging spectroscopy [22]), it is not a priori clear whether the OPA procedure
should be performed in such a way as to obtain initial estimates of signal decay curves
(according to ref. [15]) or to provide initial estimates of NMR spectra. Secondly, the
constraints that can be imposed to the C (signal decays) and S (NMR spectra) matrices are
of different nature. As mentioned above, it is assumed that the signal decay of both
components should show some kind of exponential behaviour. The possibility of fitting
each column of the C matrix to an exponential model can be imposed as a constraint within
the ALS procedure. This has been called in the literature as hard-soft modelling [23].
In this article we aim to describe and assess the different configurations of the OPA-ALS
method as applied to solid-state NMR data. The different possibilities of applying OPA
(obtaining initial estimates for NMR peaks or for exponential decays) are critically
evaluated. Also, different configurations of the ALS method can be used. Some of the
results obtained are significantly different when considering the different configurations.
2. Theory
2.1. Hahn-Echo decay in solid-state 1H-NMR of a polymer network.
A T2 Hahn-Echo Pulse Sequence (HEPS) relaxation experiment in 1H-NMR can be
classified as a second-order experiment, yielding a second-order tensor of data (i.e. a
matrix). The first order of measurement is the NMR-frequency axis expressed in relative or
absolute frequency units (ppm or kHz), whereas the second order is related to the relaxation
time (measured in time units). Following the terminology in Eq. (1), D contains the
experimental data obtained from a T2 HEPS experiment. D is n × m sized, n being the
number of relaxation times assayed, and m the number of frequency points collected in the
NMR spectrum. An arbitrary element of this matrix, di,j, collects the signal observed at the
ith pulse spacing (t) and the jth frequency. Accordingly, a column in the D matrix, dj, is a
Appendix 1
130
column vector (nx1 sized) representing the signal decay (intensity measured at different t
values) at the jth frequency (or chemical shift). Similarly, a row in the D matrix, di, is a row
vector (1xm sized) showing an NMR peak profile (intensity at different frequencies) at the
ith pulse spacing.
When applying the HEPS technique to a polymer network, different parts of the network
may produce different relaxation decays and NMR spectra. All these different molecular
parts (or “components”) are characterized, on the one hand, by different mobilities (different
relaxation behaviour) and, on the other, by different chemical shifts (different NMR peaks
or peak shapes). A single component of the network will give rise to a unique NMR peak
and relaxation behaviour, both profiles being independent of each other. In other words, the
relaxation behaviour of a given component is independent to the frequency at which it is
measured and the NMR spectrum has the same shape at any relaxation time (see Appendix
for a more-detailed justification of this assumption). This implies, strictly speaking, that the
D matrix is bilinear, so we can assume the model described in Eq. (1). For p components
present, Eq. (1) hence adopts the form:
(2) εDεCSεscDp
1kkk
ˆ
where ck is a column vector (n×1 sized) of the signal decay for component k, and sk the row
vector (1×m sized) containing the NMR spectrum for component k, and a matrix (n×m
sized) collecting the error of the model. Therefore, C is a matrix of relaxation decays (n×p
sized) formed adjoining all the ck elements. Similarly, S is a matrix of NMR spectra (p×m
sized) formed adjoining all the sk elements. The hat symbol (“^”) is used to indicate the part
of the experimental matrix that is modelled: . CSD ˆ
1H NMR Hahn-Echo decays of cross-linked elastomers can often be empirically described
in terms of two or three exponentials [16-18]. The fastest decay component typically arises
from short network chains whose mobility is strongly restricted by crosslinks at both sides.
The slower decay component(s) are due to longer network chains, so-called “dangling ends”
connected to the network at just one end of the chain, and a sol fraction of uncured material.
In practice it is often assumed that every decay component adopts an exponential form in
the relaxation decay. Therefore, for a two-component model (p=2) the ith-row and jth-
column element of the D matrix , is defined as: ˆjid ,
ˆ
Appendix 1
131
biai Ttj
Ttjji ebead 22
,ˆ (3)
where the terms aj and bj are the respective intensities of components A and B at the jth
value of the chemical shift along the relative frequency axis, ti is the ith echo time in the
HEPS experiment, and T2a and T2b are the respective transversal-relaxation times of the two
components. If the last equation holds, matrix C is defined as follows:
bnan
ba
ba
TtTt
TtTt
TtTt
22
2222
2121
expexp
......
expexp
expexp
C (4)
and matrix S is defined as:
(5)
m
m
bbb
aaa
...
...
21
21S
where a1, a2, ... am and b1, b2, ... bm are defined in Eq. (3).
2.3. The OPA-ALS method
2.3.1. ALS1, ALS2, ALS3 and ALS4
The task of solving Eq. (1) consists of finding the C and S pair of matrices that minimise the
differences between D and CS. This implies forcing the unmodelled variance () to be
minimal. Mathematically speaking, this can be formulated as minimising the Sum of
Squares of Residuals:
(6)
m
j
n
ijiSSR
1 1
2,
22 εCSD
Unfortunately, there is more than one pair of CS matrices that make SSR minimal. For any
solution C, S = C0, S0 of Eq. 1, there is a continuous family of mathematically correct
solutions C0 T-1, T S0, where T is any invertible (pp) matrix. It is obvious that the
problem has an infinite number of solutions, because an infinite number of T matrices exist.
This is called rotational ambiguity.
Additional constraints on C and S can offer a way out of this rotational ambiguity. The ALS
algorithm is constructed in such a way that a solution of T is varied in order to direct the C
Appendix 1
132
and S matrices to a final result that meets certain (minimal) requirements for a physically
meaningful solution. Specific constraints are imposed on C and S (e.g. unimodality or non-
negativity [9]) depending on the nature of the problem. In the problem presented here,
different constraints are applied to C and S. As explained before (see section 2.1. and
Attachment), it is assumed that the columns in the C matrix are described by exponential
decays [24]. An additional reasonable assumption is that the NMR spectra in matrix S are
positive across the whole frequency range, so that a non-negativity constraint is applied to
S. In the ALS algorithm these constraints are used to direct the process to the true or, at
least, a physically meaningful solution.
However, ALS can be applied in two different ways, denoted as ALS1 and ALS2 in Figure
A1.1. Antecedents of these two configurations can be found in ref. [25]. In ALS1 we start
with an approximation of the C matrix, C0, in which the subindex 0 denotes that this is the
0th iteration. In a second step, constraints are applied to C0. The points in each column of the
C0 matrix are fitted to an exponential decay. This can be done for each column
independently by linear regression, after calculating the logarithm of the values. The fitted
points (once retransformed back to the exponential form) constitute the constrained matrix,
, where the superindex c indicates that the constraints have been applied. In the linear
regression, not only is obtained, but also the values of T2a and T2b (cf. Eq. (3), Figure
A1.1). If is known, an estimate of S0 can be obtained by minimizing SSR:
c0C
c0C
c0C
DCS
c00 (8)
where denotes the Moore-Penrose generalised inverse of . In a subsequent step,
constraints are applied over S0 (non-negativity) zeroing those values from S0 that are
negative. This yields a constrained matrix of NMR peak profiles, . In a way similar to
Eq. (8), a new estimate for C (C1) can be found if is known:
c0C
c0C
Sc0
c0S
c01 SDC (9)
All these operations constitute a closed loop, so the whole process can be repeated using
subsequnt estimates of C (Ck) until convergence is reached (see Figure A1.1). The
convergence is checked by monitoring SSR, which is calculated in this case by substituting
Appendix 1
133
C1 and in Eq. (6). When SSR starts to increase or no longer decreases significantly, the
process is stopped. Note that when the convergence is reached after q iterations, a matrix of
unconstrained signal decays (Cq+1) and constrained NMR peaks ( ) is obtained.
c0S
cqS
A second way in which the ALS algorithm can be applied is denoted by ALS2 in Figure
A1.1. We now start with an approximation of the S matrix, S0. From S0 a constrained
matrix, , is obtained after applying constraints (non-negativity). Eq. (9) is then used to
find the unconstrained C matrix, which we call C0 instead of C1 in this case. This matrix is
converted into after applying constraints (fitting exponential decays to the columns of
C0 and obtaining T2a and T2b). Eq. (8) then yields to a new solution of matrix S, S1. The
process can be repeated until convergence is reached. In this case SSR is calculated using S1
and . A matrix of unconstrained NMR peaks ( ) and a matrix of constrained (fitted)
signal decays ( ) are obtained after q iterations. Note that ALS1 and ALS2 do not differ
only in the initial guesses (column-wise or row-wise), but also in the quantity used to
monitor convergence (SSR is calculated differently) and type of outputs given
(unconstrained or constrained).
c0S
c0C
cqC
c0C 1qS
A deeper examination of the algorithm's architecture reveals the existence of two extra
alternatives, both based on the way in which SSR is calculated. One can keep in memory
the regressed (unconstrained) solutions of Sq and Cq throughout the loop, and use their
product to calculate and monitor SSR. The logical output of the algorithm is then a
completely unconstrained solution. We have labelled this procedure as ALS3 (see Figure
A1.1). An alternative consists of keeping in memory the constrained solutions, and
that have been obtained during a single loop, and use them to calculate SSR. The output of
the algorithm, which will be called ALS4, is then a constrained collection of exponential
decays and NMR profiles.
cqS
cqC
2.3.2. OPA1 and OPA2
From the last section, it is obvious that an initial estimate of the C or the S matrix is needed
to start the ALS1 or the ALS2 algorithm, respectively. This can be solved in several ways,
the most popular of which is the Orthogonal-Projection Approach (OPA [7]). Only a brief
description of the OPA method is given here. More details can be found elsewhere [7].
Appendix 1
134
Figure A1.1. Flow diagram of the algorithms OPA1-ALS1,OPA1-ALS2, OPA2-ALS1 and OPA2-ALS2.
Appendix 1
135
Let us suppose that we want to obtain an initial estimate of the C matrix (signal decays).
The OPA method is based on finding those columns in the D matrix that give the most
dissimilar signal decays. For that purpose, the dissimilarity at each chemical shift, dj, is
calculated:
jT
jjd XXdet (10)
where the Xj matrix is defined by adjoining a set of reference-signal decays, R, with the
signal decay at the jth chemical shift:
jdRX (11)
where dj is the vector containing the jth column of D. In the first iteration, the R matrix
contains the mean signal decay (taken across all the signal decays of D). Once the chemical
shift with the maximum dissimilarity has been located (the ath row in D), the corresponding
signal decay at this particular chemical shift, da, replaces the mean in R: . After this
first iteration, the dissimilarity is calculated again across all signal decays and the chemical
shift with maximum dissimilarity is located (the bth row in D). The R matrix then
incorporates db, increasing its number of columns:
adR
bdRR . This operation is repeated
until the number of columns of the R matrix equals the number of expected components
(which is two in our example). Once the process has ended, the resulting R matrix is taken
as the initial estimate of the C matrix.
However, the OPA method can also be applied to obtain an estimate of the S matrix. In this
case, Eq. (11) takes the form:
TidRX (12)
where diT is the transposed vector containing the ith row of D. In the same way, in the first
iteration the R matrix contains the transposed mean NMR spectrum (taken across all the
NMR spectra contained in D). In this case, the method is based on finding those time delays
that yield the most dissimilar NMR spectra. In each iteration, the most dissimilar transposed
NMR spectrum is incorporated in the R matrix as a new column. When the number of
columns of R equals the number of expected components (two in our case), the transpose of
R is taken as the initial estimate of S. We will define OPA1 and OPA2 as the methods
yielding initial estimates of S and C, respectively.
Appendix 1
136
other order of measurement. When a
At this point, applying OPA1 before ALS1 or OPA2 before ALS2 are straightforward
ms OPA1-ALS1 and OPA2-ALS2. However, two other
The samples were supplied by DSM Resins (Zwolle, The Netherlands) in the form of
ter resins in 33 wt % of styrene. A catalyst (0.5% of a 1%
phthalic anhydride with 1,2-propylene glycol and ethylene glycol. The material was brought
The OPA method is usually applied to the order of measurement with the largest selectivity,
subsequently comparing dissimilarities in the
chromatographic technique is involved (e.g., HPLC-DAD), the selectivity is larger in the
chromatographic direction (chromatograms) than in the spectral direction (spectra). It is
typically easier to find elution zones in which only one component is eluting, than to find
wavelengths at which only one component is absorbing. Consequently, the OPA method is
normally applied to obtain estimates of the spectra of the components, rather than of the
chromatographic peak profiles. However, when the instrument does not involve a
chromatographic separation (as in the example studied in this paper), it is not clear which
order of measurement yields the maximum selectivity. So far, to the authors’ knowledge, no
studies have been performed to compare the results of processes starting with OPA-based
estimates of C or S in NMR techniques.
2.3.3. Eight ALS configurations
options. We will call these algorith
combinations are also possible. After applying OPA1 to obtain an estimate of C, Eq. (8) can
be applied (without applying constraints to C) to obtain an estimate of S and the ALS2
configuration can subsequently be applied (OPA1-ALS2). In a similar way, OPA2-ALS1
can be considered by applying Eq. (9) to matrix S to obtain an initial estimate of C, before
starting ALS1. Finally, ALS3 and ALS4 can also be applied with both OPA1 and OPA2. In
total, eight configurations are possible, namely: OPA1-ALS1; OPA1-ALS2; OPA1-ALS3;
OPA1-ALS4; OPA2-ALS1; OPA2-ALS2; OPA2-ALS3; OPA2-ALS4;
3. Experimental
3.1. Sample preparation
solutions of unsaturated-polyes
cobalt solution) and 2% of benzoyl peroxide initiator (2%) were added to the solution. The
polyester chains had been formed by a polycondensation reaction of maleic anhydride and
Appendix 1
137
00 spectrometer equipped with
a 7-mm Magic Angle Spinning (MAS) probe. To prevent unwanted sample rotation caused
ting, an empty sample holder was loaded on top of the one
AB 7 (The Mathworks, Natick, MA, USA), were
used for data treatment.
4.1. Comparison of OPA1 and OPA2 to obtain initial estimates
are presented in Figure A1.2 and Figure A1.3),
respectively, and correspond to the HEPS-NMR experiments at 505 K. Results obtained at
raphs a-c (in
between two glass plates, degassed and then cured for 24 hours at room temperature.
Thereafter, two post-curing step were performed during 24 hours each at 600C and 800C,
respectively. For the NMR experiments, the resulting networks were ground to a fine and
apparently homogeneous powder using a cryogenic grinder.
3.2. NMR experiments
The NMR experiments were performed on a Bruker DMX5
by the N2 flow used for hea
containing the sample material. The relaxation experiments reported in this article were
carried out at 300, 441, 465 and 505 (calibrated temperature scale) on the cross-linked
samples. The Hahn-Echo pulse 90º--180- sequence (HEPS) with 180-pulse duration 10
s and echo times 2 between 0.004 and 4 ms (see section 2.1.) was used to record the
decay of the transverse magnetization.
3.3. Software
Home-built routines, written in MATL
4. Results and discussion
The results of OPA1 and OPA2
other temperatures were similar and these are not presented here for brevity. G
both figures) depict the R matrix when zero (part a), one (part b) and two (part c)
components are considered (see figure captions for a more detailed explanation). As can be
seen, OPA2 yields better initial estimates when compared with OPA1. This will be
confirmed numerically in the next section. At this point, the result can be tested qualitatively
by comparing the vector of dissimilarities (part f) in both figures. In Figure A1.2f two bands
of high dissimilarity are clearly observed at both sides of the center of the NMR spectrum.
Appendix 1
138
mpares the results in terms of the Square Sum of Residuals -SSR; Equation
(6)- obtained when applying the eight configurations to data obtained at four different
n the
nt. ALS1 yields unconstrained relaxation decays and constrained NMR
This means that the initial (signal decay) estimates (matrix R, Figure A1.2c) are not
sufficient to explain all the variance of the signal. In other words, when the columns of the
D matrix cannot be explained as a linear combination of the vectors contained in R, the
determinant of Eq. (10) increases. This is because the most dissimilar signal decay is always
found around the NMR peak maxima (Figures A1.2d and A1.2f). In contrast, Figure A1.3f
does not show equally clear (highly dissimilar) zones as Figure A1.2f. Although one
dissimilar zone is found around 0.01 ms, the general trend is noisier than Figure A1.2f. This
is because the most dissimilar regions in parts 2d and 2e are found at different time decays,
yielding thus less-correlated NMR spectra (Figure A1.2c). The differences in quality of the
initial estimates will be even clearer in the next section, in which the numerical results will
be compared.
4.2. Comparison of the results obtained with the eight configurations
Table A1.1 co
temperatures. The results that were qualitatively evaluated in the previous section o
performance of OPA1 and OPA2 are confirmed here quantitatively. For the same kind of
output, OPA2 always gives rise to lower SSR values. On the other hand, OPA1 caused most
of the divergence situations. Therefore, it can be concluded that the initial estimates
obtained by evaluating the dissimilarities between signal decays (i.e. OPA1) give rise to a
local minimum.
The results comparing the different ALS versions (using the same input) can be interpreted
considering the fact that outputs obtained from ALS1, ALS2, ALS3 and ALS4 algorithms
are not equivale
spectra, while ALS2 yields constrained relaxation decays and unconstrained NMR
spectraALS3 yields both unconstrained results, whereas in ALS4 both orders of
measurement are constrained. Any constraint implies an increase in SSR (the function has
less flexibility to adapt to the experimental data). Fitting an exponential decay (hard-
modelling) implies a stricter constraint than applying non-negativity (soft-modelling).
Obviously, the stricter the constraints are, the higher the SSR values will be. Thus, lower
values of SSR are expected when using ALS1 compared to the results obtained with ALS2.
Appendix 1
139
Figu a through through e the diss an be followed o calculate the id in this
cay is located. T idering th
d and e are repr ese two com
res A1.2. Illustration of the OPA1 method for the NMR data at T = 505 K. Parts c depict initial guesses (R matrix of signal decays), whereas in parts dimilarity curves are represented (NMR spectra). The OPA procedure c from left to right and from top to bottom. Part a depicts the mean decay used t
dissimilarity curve (part d). The maximum of the curve is overlagraph (dashed line), which indicates the frequency at which the highest dissimilar de
his decay is depicted in part b. Part e depicts the dissimilarity curve conse decay represented in part b as initial guess. The maximum of the dissimilarity is again
overlaid as a dashed line. The two extracted decays of maximum dissimilarity in parts esented together in part c (solid and dashed lines). The dissimilarity with th
ponents is depicted in f. All y-axes are relative.
Appendix 1
140
ive.
Figure A1.3. Illustration of the OPA2 procedure for the NMR data at T = 505 K. Similarly to Fig. A.2, parts a through c depict initial guesses (NMR spectra), whereas in parts d through e the dissimilarity curves are represented (decays). The procedure of OPA can be followed from left to right and from top to bottom. Part a depicts the mean spectrum used to calculate the (dis-)similarity curve (part d). The maximum of the curve is overlaid in this graph (dashed line), indicating the time decay at which the most dissimilar spectrum is located. This spectrum is depicted in part b. Part e depicts the dissimilarity curve considering the spectrum represented in part b as initial guess. The maximum of the dissimilarity is again overlaid as a dashed line. The two extracted spectra of maximum dissimilarity in parts d and e are represented together in part c (solid and dashed lines). The dissimilarity with these two components is depicted in part f. All y-axes are relat
Appendix 1
141
Also, ALS3 yields normally the lowest SSR value (when compared with other algorithm
configurations using the same input), whereas ALS4 gives normally the highest. This is
clear from Table A1.1 with OPA2. Exceptions to this rule are found in case of divergence,
due to the different ways in which SSR is calculated to monitor convergence.
Method Temperature, K OPA1-ALS1 OPA1-ALS2 OPA2-ALS1 OPA2-ALS2
300 1.06 7.55 0.47 1.91
441 0.18 1.09 0.17 2.22
465 5.60 10.75 0.32 3.39
505 1.50 2.60 0.16 0.86
Figure A1.4 depicts the graphical results (both NMR profiles and exponential decays)
obtained with ALS1 and ALS2 (starting with both OPA1 and OPA2) at the highest
temperature (505 K). Results obtained with ALS3 and ALS4 are not included for brevity. It
can be observed that the final constrained NMR spectra, , once introduced in Eq. (9),
yield a collection of signal decays, , that do not meet the constraints. This can be seen
clearly in Figure A1.4a and Figure A1.4e, in which th put from OPA1-ALS1 is
depicted. One of the unconstrained signal decays is by no means an exponential decay. A
similar effect – be it to a le hen using the OPA2-ALS1 configuration
to
the signal decays obtained with ALS1, whereas the NMR profiles are similar to the NMR
Table A1.1. SSR (×1017) obtained with the eight configurations explained in the text.
cfS
e out
fC
sser extent – is found w
(Figure A1.4c and A1.4g). Similarly, but less clearly, the final constrained signal decays,
cfC , once introduced in Eq. (9), can yield a collection of NMR spectra, fS , that
occasionally do not meet the constrains. This effect is clearly noticeable in the results
obtained using the OPA1-ALS2 configuration (Figure A1.4b and Figure A1.4f). Forcing the
signal decays to be exponential yields negative regions in one of the NMR spectra. Starting
with more correct initial candidates (OPA2) greatly alleviates this problem. Negative parts
in the NMR spectra in Figure A1.4d are almost unnoticeable.
Similar effects are observed with ALS3 and ALS4. With ALS3, signal decays are similar
Appendix 1
142
straints, since
both signal decays and NMR profiles are always constrained. As the results obtained with
ALS3 and ALS4 can be deduced from ALS1 and ALS2, ALS3 and ALS4 configurations
tw results obtained from ALS1, ALS2, ALS3 and ALS4 are
rpreted as a la b that nen ation
too simple plain the obtained in HEPS-NMR exper s on polymer
rks [21]. H r, includin component e model led to divergence, due to
ilities in th putation of the Moore-Penrose pseudoinverse (8) and (9)-.
bly, the pea iles in NM d the signal decays were too simi r the different
omponents and the and matrices were nearly singular. Another source of a lack of fit
dis
on the scale of the signal.
profiles obtained with ALS2. Therefore, with OPA1-ALS3, both profiles (signal decays and
NMR spectra) are away from meeting the constraints. With OPA2-ALS3, the problem is
alleviated for non-negativity applied to the NMR profiles. Inspecting the results with ALS4
gives no information about how difficult is for this system to meet the con
will not be studied further.
Differences be
inte
een the
ck of fit. It has een suggested the two-compo t model of Equ
(3) is to ex results iment
netwo oweve g more s in th
instab e com -Eqs.
Proba k prof R an lar fo
S Cc
can be that the signal decays do not follow an exponential behaviour. It has been reported in
the literature [21] that the signal decay in HEPS-NMR is a combined function of Gaussian
and exponential components.
4.3. Comparing dissimilarities in both orders of measurement
A more detailed study into why the OPA2 configuration performs better than the OPA1 in
providing initial estimates is shown in Table A1.2. In this table the dissimilarities between
the deconvolved signal decay profiles and the NMR spectra profiles are compared. The
objective of this comparison is to check which order of measurement features the higher
similarity, in order to study quantitatively which order of measurement is most
appropriate for applying the OPA method. For this purpose the true solutions of C and S
should be compared. The true NMR spectra of the two components are unknown, so the
best solutions (those obtained with OPA2 configuration) have been used for the
computations. In this particular case, calculation of the dissimilarity using Eq. (10) is not
recommended, since the value of the measurement depends
Appendix 1
143
Figures A1.4. Results of a two-component decomposition after applying the four algorithms described in this article to the NMR experiments at T = 505 K. Parts a through e represent NMR peak profiles, whereas parts f through j depict time decays. Algorithms: a,e OPA1-ALS1; b,f OPA1-ALS2; c,g OPA2-ALS1; d,h OPA2-ALS2.
Appendix 1
144
ormalised dissimilarity measurements are more suitable in this case. Instead, we have
calculated the dissimilarity according to ref. [26], as the sine of the angle between the two
vectors (representing the NMR spectra or the signal decays). The resulting values are
normalized. A value of 1 implies that the profiles completely different, whereas 0 implies
that the profiles are fully equivalent.
In studies involving second-order NMR data (such as DOSY-NMR [15]), in which a series
of NMR spectra are obtained applying different time delays, a good initial guess for ALS is
found by checking those frequencies of the NMR spectrum that yield the most dissimilar
time decay. This means that the OPA1 configuration is used and that OPA2 is normally not
applied. In second-order NMR spectroscopy, more selectivity is expected in the NMR
spectra rather than in the time delay. Due to the exponential nature of the signal decays, it is
easier to find frequencies containing a single (pure) compound than specific time delays
yielding NMR spectra of a pure component. However, solid-state NMR yields a different
perspective than liquid NMR. When applying Hahn-Echo techniques on cross-linked
polymer networks, regions of selectivity (pure or dominant components) are hard to find in
either order of measurement. If only two components are considered, the maxima in the
NMR spectra of both components are found at the same frequency; only differences in band
broadening can be found. This makes the dissimilarity in NMR profiles in this particular
case always significantly lower than the dissimilarity in signal decays, independent of the
ALS method applied and on the temperature (see Table A1.2). This explains why in this
particular case OPA2 performs better.
Table A1.2. Dissimilarity obtained with the deconvolved profiles (P and D matrices) of OPA2-based methods. The dissimilarity is calculated according to ref. [23].
NMR profiles Signal decay profiles
N
Temperature,
K OPA2-ALS1 OPA2-ALS2 OPA2-ALS1 OPA2-ALS2
300 0.535 0.536 0.834 0.760
441 0.413 0.480 0.716 0.628
465 0.391 0.426 0.638 0.595
505 0.337 0.407 0.636 0.562
Appendix 1
145
each specific chemical or physical component yields a
single source of variance it is possible to separate the contributions of each component to
onstrained signal decays and NMR spectra). The
ombination of all these configurations gives rise to eight different ways of applying OPA-
As for the signals obtained from solid-state HEPS-NMR experiments on polymer networks,
a d to perform better ions. This can be
lained from a ity ec sign e only
ences in p e (and not in peak positi ed i MR spectra
corresponding to the different components of the network. A way of testing which OPA
a ch performs consists of comparing the dissimilarity of either
spectra or ignal deca rder of m ent yieldin ost dissimilar
indicates t itable O nfiguratio s been dem ed that signal
4. Conclusions
Multivariate Curve Resolution (MCR) methods are designed to interpret data from second-
order instruments. These methods are based on the separation of the sources of variance that
are contained in the output matrix. If
the raw experimental data. Solid-state HEPS-NMR experiments on polymer networks yield
second-order data, which can be analysed using MCR methods. If the presence of two
components is assumed, the variance produced by each component can be determined, and
the contribution of each component to the total signal can be established. Orthogonal-
Projection Approach in combination with Alternating Least Squares (OPA-ALS) belongs to
the family of MCR methods. It has been applied in this work to analyse and separate the
contributions of two components present in polyester polymer networks. Each component is
attributed to a part of the polymer with a different mobility.
The OPA method can be applied in two ways, i.e. by evaluating the dissimilarities in signal
decays at each frequency in the NMR spectra (OPA1) or by evaluating the dissimilarities in
NMR spectra at each time delay (OPA2). The ALS method can also be applied in four ways
(ALS1, ALS2, ALS3 and ALS4), depending on (i) which order of measurement is used as
initial estimate (signal decays or NMR profiles) (ii) which profiles (unconstrained or
constrained) are used to calculate SSR to monitor convergence and (iii) the output produced
by the algorithm (constrained or unc
c
ALS. All of these have been compared in this work, yielding different results.
OPA2 configur
exp
tions were foun than OPA1 configurat
lower selectiv in the NMR sp tra than in the al decays, sinc
differ eak shap on) are expect n the N
pproa best the deconvolved
NMR of the s ys: the o easurem g the m
value he most su PA co n. It ha onstrat
Appendix 1
146
milar than NMR spectra, so that OPA2 is recommended instead of
OPA1.
decays are more dissi
Differences between the outputs obtained from the different ALS configurations and the
same OPA strategy have to be interpreted with caution. As the different configurations force
the output to be more or less constrained, the sum of squared residuals (SSR) obtained with
this output cannot be compared directly. On the other hand, significant differences were
observed between the retrieved spectra and signal decays from the different ALS
configurations. This can be interpreted either as a lack of fit or rotational ambiguity. Among
all configurations, OPA2-ALS2 performed the best and its use is recommended for the
analysis of solid-state HEPS-NMR data obtained on polymer networks.
Appendix 1
147
[1] Booksh, K. S.; Kowalski, B. R. Anal. Chem. 1994, 66, 782A.
[9] de Juan, A.; Tauler, R. Anal. Chim. Acta 2003, 500, 195.
0] Wiberg, K. J. Chromatogr. A. 2006, 1108, 50.
[11] Li, H.; Zhang, F.; Havel, J. Electrophoresis 2003, 24, 3107.
[12] Garrido, M. ; Larrechi, M. S. ; Rius, F. X. ; Tauler, R. Chemom. Intell. Lab. Syst. 2005, 76, 111.
[13] Wang, J. H. ; Hopke, P. K. ; Hancewicz, T. M. ; Zhang, S.L. Anal. Chim. Acta 2003, 476, 93.
[14] Bezemer, E. ; Rutan, S. Anal. Chim. Acta 2002, 459, 277.
[15] Huo, R. ; Wehrens, R.; van Duynhoven, J.; Buydens, L. M. C. Anal. Chim. Acta 2003, 490, 231.
[16] Litvinov, V. M. ; Dias, A. A. Macromolecules 1999, 32, 4051.
[17] Litvinov, V. M. ; Barthel, H. ; Weis, J. Macromolecules 2002, 35, 4356.
[18] Orza, R. A. ; Magusin, P. C. M. M.; Litvinov, V. M.; van Duin, M.; Michels, M. A. J. Macromolecules 2007, 40, 8999.
[19] Cohen-Addad, J. P.; Dupeyre, R. Polymer 1983, 24, 400.
[20] Simon, G.; Baumann, K.; Gronski, W. Macromol. 1992, 25, 3624.
[21] Saalwachter, K. Macromolecules 2005, 38, 1508.
[22] Duponchel, L.; Elmi-Rayaleh, W.; Ruckebusch, C.; Huvenne, J. P. J. Chem. Inf. Comput. Sci. 2003, 43, 2057.
[23] de Juan, A.; Maeder, M.; Martínez, M.; Tauler, R. Chemom. Intel. Lab. Syst. 2000, 54, 123.
[24] Huo, R.; Wehrens, R.; Buydens, L. M. C. J. Magnetic Res. 2004, 169, 257.
[25] Gargallo, R.; Tauler, R.; Cuesta-Sánchez, F.; Massart, D. L. Trends in Analytical Chemistry 1996, 15, 279.
[26] Vivó-Truyols, G.; Torres-Lapasió, J. R.; van Nederkassel, A. M.; Vander Heyden, Y.; Massart, D. L. J. Chromatogr. A. 2005, 1096, 146.
References
[2] Mason, C.; Maeder, M.; Whitson, A. Anal. Chem. 2001, 73, 1587.
[3] Tauler, R.; Barcelo, D. Trends Anal. Chem. 1993, 12, 319.
[4] Kvalheim, O.M.; Liang, Y. Z. Anal. Chem. 1992, 64, 936.
[5] Fraga, C. G.; Bruckner, C. A.; Synovec, R. E. Anal. Chem. 2001, 73, 675.
[6] Faber, N. M.; Bro, R.; Hopke, P. K. Chemom. Intel. Lab. Syst. 2003, 65, 119.
[7] Cuesta-Sánchez, F.; Toft, J.; van den Bogaert, B; Massart, D. L. Anal. Chem. 1996, 68, 79.
[8] Windig, W.; Guilment, J. Anal. Chem. 1991, 63, 1425.
[1
Appendix 2
149
Appendix 2
Combining linear and non-linear regression algorithms as an alternative
for multivariate curve resolution problems of low selectivity
1. Introduction
Consider solving the following equation for bilinear data:
(1) εYεPDY ˆ
where Y stands for the experimental data generated by a second-order instrument [1], P and
D are the matrices containing (column-wise and row-wise, respectively) the profiles of each
chemical component in the first- and second-order of measurement, is de unmodelled
variance and by definition PDY ˆ . If the instrument is, for example, a HPLC-DAD, then
Y is the spectrochromatogram of a mixture, and P and D contain the chromatographic peak
profiles and the spectra of each chemical component, respectively.
As normally Y is (experimentally) known, the chemical problem consists of retrieving P
and D, i.e., the profiles of each chemical component in both orders of measurement. This
can be done in different ways [2], most of them based on finding the least-squares solution
that minimizes . Unfortunately, there is not a unique PD pair of matrices that minimizes .
This problem has been formulated as "rotational ambiguity" [3] and arises from considering
a rotation matrix T (T'T=I) that:
TDPT'PT (2)
where Pn=PT' and Dn=TD can be considered as a new pair of PD matrices that satisfy Eq.
(1). As an infinite number of T matrices exist, an infinite number of solutions exist for Eq.
(1).
Appendix 2
150
There are two ways to constrain the number of solutions of Eq. (1). The first one involves
considering more than one experiment (i.e., more than one Y matrix), building up three-way
data. The most popular methods of this family are, among others, PARAFAC [4] or GRAM
[5]. An alternative, when three-way data is not available -or its use is not convenient-, is to
make use of chemical knowledge. From the chemical point of view, only certain P and D
profiles make sense. For example, chromatographic peak profiles (in P matrix) should be
unimodal and non-negative, and spectra (in D matrix) should be non-negative. These
constrains can be imposed in P and D to restrict the number of solutions. The family of
techniques that impose constraints on P and D based on chemical knowledge was
introduced some decades ago by Tauler et. al., and -having many variants- has been named
as Multivariate Curve Resolution (MCR).
Two kinds of constraints can be distinguished in MCR methods. The first has been named
soft modelling [6], and does not fit the data to a mathematical equation, but corrects the P
and D matrices to meet the constraints. A typical example in the field of HPLC-DAD is to
correct for the negative parts of all spectra contained in D matrix, zeroing the values that are
below zero. An alternative (called hard modelling) is to fit the data to a mathematical
equation. Following the example before, the chromatographic profiles contained in P could
be fitted to a peak model.
The present work is related to this second sub-family of techniques. Traditionally, hard
modelling is applied as an extra constrain inside the MCR algorithm, so the fitting of the
mathematical equation is nested as an extra operation inside the MCR process. This
alternative was studied in a previous article [7]. It was found that several configurations of
the algorithm were possible, and all of them were studied. Another alternative is presented
and explored in this work. It consists of fitting the model parameters with a non-linear
algorithm (which gives an estimation of P, for example), and nesting a linear fitting (to
calculate D in this case) inside. Both methods seem to be complementary, having
advantages and disadvantages. All of them are analyzed in this work.
The example used to illustrate the algorithm performance is a solid-state Hahn-echo NMR
data of a polymer network. With this technique, changing the delay time between NMR
excitation and signal measurement, second-order data is obtained: one of the order of
measurement accounts for conventional chemical shifts (NMR spectrum) and the other for
Appendix 2
151
the signal decay. It can be assumed that the signal decay follows an exponential behavior,
which brings the possibility of fitting each column of the P matrix (each component of the
polymer network) to an exponential model (hard modeling). A direct outcome of the
algorithm is an estimation of the relaxation time of the components of the polymer network,
which is highly valuable to characterize the polymer topology.
2. Experimental
2.1. Sample preparation
The samples were supplied by DSM Resins (Zwolle, The Netherlands) in the form of
solutions of unsaturated-polyester resins in 33 wt % of styrene. A catalyst (0.5% of a
1% cobalt solution) and 2% of benzoyl peroxide initiator (2%) were added to the
solution. The polyester chains had been formed by a polycondensation reaction of
maleic anhydride and phthalic anhydride with 1-2 propylene glycol and ethylene glycol.
The material was brought between two glass plates, degassed and then cured for 24
hours at room temperature. Thereafter, two post-curing step were performed during 24
hours each at 600C and 800C, respectively. For the NMR experiments, the resulting
networks were ground to a fine and apparently homogeneous powder using a cryogenic
grinder.
2.2. NMR experiments
The NMR experiments were performed on a Bruker (Karsruhe, Germany) NMR
spectrometer of 500 MHz. The relaxation experiments were carried out at different
temperatures (from 213 K to 450 K) on the cross-linked samples. The Hahn-echo pulse
sequence (HEPS) measurement consists of a double-pulse sequence: 90°x−tHe –
180°x−tHe—acquisition, where tHe is varied between 1 and 4 ms. The second pulse in the
HEPS inverts nuclear spins of mobile molecules only, and an echo signal is formed with
a maximum at time t =2 tHe after the first pulse. By varying the pulse spacing in the
HEPS (t), information about intensity profiles is obtained as a function of time t. A
single t2 HEPS relaxation experiment in 1H-NMR yields thus two-way data. The first
order (measured in frequency units) accounts for conventional chemical shifts, whereas
Appendix 2
152
the second is related to the relaxation time (measured in time units). For consistency
with section 1, we will call Y to the experimental data-matrix obtained from a single
NMR experiment.
2.3. Solid-state NMR of a polymer network
When applying the experiment described in the previous section to a polymer network, the
relaxation signals in polymer networks will look more like different molecular parts that
produce separate relaxation signals of different form. All these different molecular parts
(or “components”) are characterized, on one hand, by different mobilities (relaxation
behaviour), and, on the other, by different chemical shifts (different NMR peaks). A
component of the molecule will give rise to a unique NMR peak and relaxation
behaviour, being both profiles independent one to each other. If this happens it is said
that the matrix Y is bilinear, so Eq. (1) holds. In Eq. (1), P is a matrix of relaxation
decays (n×p sized). Similarly, D is a matrix of NMR peaks (p×m sized). In this context,
each k “component” corresponds to each part of the molecule producing a different
relaxation decay and NMR peak. According to Litvinov’s approach [ref.], we assume
that only two components are present, so p = 2: one component (component A) refers to
the cross-links chains of the network, whereas the second component (component B)
corresponds to the fraction of more mobile components [8]. In general it is also assumed
that both components adopt an exponential form in its relaxation decay. Therefore the
ith-row and jth-column element of the matrix -see Eq. (1)-, , is defined as: Y jiy ,ˆ
biai Ttj
Ttjji ebeay 22
,ˆ (3)
where the terms aj and bj are the signals at a certain j value of the chemical shift
corresponding to components A and B, respectively; ti is the ith pulse spacing in the
HEPS experiment; and T2a and T2b are constants related to the half-time relaxation
decay for each component.
2.4. The OPA-ALS method
In a previous article the performance of several configurations of OPA-ALS algorithms was
studied [7]. Only a brief description is given here (for full details see the previous
reference). The OPA-ALS algorithm consists of two parts, namely (i) the OPA method [9]
Appendix 2
153
(retrieves initial estimates of the P or D matrix) and (ii) the ALS algorithm (which finds a
solution of Eq. (1), starting from the initial estimates found by OPA). Two versions of the
OPA method were proposed [7] (OPA1 and OPA2), depending on how the Y matrix is
analysed (column-wise or row-wise) giving an estimation of P or D in each case. The ALS
algorithm can be designed also in two ways. Taking into account that P and D matrices are
forced to meet certain constraints, ALS can give two kinds of answers: unconstrained P and
constrained D (i.e., ALS1 algorithm) or constrained P and unconstrained D (i.e., ALS2
algorithm) [7]. Therefore, four combinations of OPA-ALS algorithm exist: OPA1-ALS1,
OPA2-ALS1, OPA1-ALS2, OPA2-ALS2.
2.5. The non-linear algorithm
This algorithm arises from the idea that the (constrained) matrix P -as described in Eq.
(3)- can be described using a limited number of parameters. Taking into account that the
exponential decay of each component can be described with only one parameter, in the
example shown only two values (T2a and T2b) are needed to describe P, since only two
components are assumed. The method consists of applying a non-linear procedure to
minimise SSR. In this work the quadratically convergent Powell method [10] was used.
Starting with an initial set of (T2a and T2b) parameters, the Powell algorithm works by
varying the initial parameter estimates, evaluating the quality of the fitting simulating a
trial-and-error procedure. Each time that the quality of the fitting for a set of model
parameters is requested by the Powell routine, the SSR is evaluated -Eq. (3)-. This
implies several steps. First of all the P matrix is calculated (provided that the
parameters of the relaxation decays are supposed). In a second step, the D matrix
(containing the NMR peak profiles) is computed using Eq. (8). Once D is known, SSR
is calculated. It should be noted that Eq. (8) implies in fact a linear regression, which is
nested into a general non-linear fitting. The algorithm ends when the Powell method
does not find any significant decrease in SSR.
Appendix 2
154
3. Results and discussion
3.1 Comparison of OPA-ALS methods with non-linear method
There exist two relevant differences between the family of OPA-ALS methods
explained in ref. [7] and the non-linear method: ALS-based methods can be designed to
apply constrains in both P and D matrices, whereas in the non-linear algorithm, only
constraints on P or on D are possible. The non-linear method was designed in this
particular case to apply constraints on P, but to modify the algorithm in order to apply
constraints in D instead is straightforward. Another important difference refers to the
type of constraints that can be applied: whereas hard- and soft-modelling can be applied
to ALS, only hard-modelling is possible with the non-linear algorithm. Indeed, the
constraints applied in P (or D) matrices should consist of fitting a parametric model:
designing the non-linear algorithm to apply soft-modelling (i.e. constraints as non-
negativity or unimodality) is conceptually impossible, since the non-linear algorithm
(Powell in this case) can run only fitting parametric models. Therefore, OPA-ALS is
more flexible –in terms of constraints– than the non-linear algorithm. In this particular
case the non-linear algorithm is however specially suited: hard-modelling is naturally
applied (an exponential decay is fitted to P), and –as will be seen– to apply constraints
on D is not mandatory.
In order to evaluate properly how the non-linear algorithm and the OPA-ALS methods
compete, in a first experiment all OPA-ALS algorithms were run without applying any
constraint in D. The influence of applying non-negativity to the D matrix will be
discussed later. The results of the five methods (OPA1-ALS1, OPA1-ALS2, OPA2-
ALS1, OPA2-ALS2 and non-linear algorithm) are presented in Tables A2.1, A2.2 and
A2.3, respectively. As can be seen attending to the residuals, the OPA1-ALS2 (and in a
lesser extent OPA2-ALS2) method yields the lowest residual. This is not surprising,
since one should take into account that the ALS2 method yields unconstrained P and
constrained D. Moreover, in this particular case no constraints were applied to D,
therefore the results given by ALS2 methods are completely unconstrained. The more
the constrained the solution is, the higher the SSR will be, since the modeled function
has less flexibility to adapt to the experimental data. Therefore, the comparison of ALS2
with ALS1 or non-linear algorithm is not fair.
Appendix 2
155
Table A2.1. SSR (×1017) obtained with the five algorithms explained in the text. No constraint was applied to D.
Temperature OPA1-ALS1
OPA1-ALS2
OPA2-ALS1
OPA2-ALS2 non-linear
1 1.91 0.46 7.67 0.47 1.31
2 2.06 0.17 9.29 0.18 0.88
3 2.92 0.32 23.83 0.36 1.54
4 0.66 0.16 5.78 0.17 0.45
Temperature OPA1-ALS1
OPA1-ALS2
OPA2-ALS1
OPA2-ALS2 non-linear
Table A2.2. T2a values obtained with the five algorithms explained in the text. No constraint was applied to D.
1 0.011 0.011 0.075 0.312 0.010
2 0.044 0.033 0.464 0.495 0.051
3 0.076 0.054 0.670 0.725 0.084
4 0.109 0.073 0.687 0.743 0.109
Table A2.3. T2b values obtained with the five algorithms explained in the text. No constraint was applied to D.
Temperature OPA1-ALS1
OPA1-ALS2
OPA2-ALS1
OPA2-ALS2 non-linear
1 0.482 0.482 0.015 0.040 0.064
2 0.899 0.922 0.202 0.154 0.564
3 1.392 1.338 0.417 0.314 0.910
4 1.667 1.581 0.420 0.322 1.084
Appendix 2
156
On the contrary, ALS1 and the non-linear method can be compared properly, since both
yield to constrained P and unconstrained D. Among these three algorithms, non-linear
algorithm generated the lowest SSR. Thus, having the possibility of hard modeling, the
combination of linear and non-linear fitting performs better than ALS-based methods. It
seems that, when the non-linear algorithm governs the algorithm architecture, the model
parameters (T2a and T2b) are varied in a more smart way to get a better solution.
The solutions of the five algorithms are presented graphically in Figure A2.1 for the
highest temperature. As can be seen, only OPA1 algorithms and non-linear algorithm
give reasonable results on NMR spectra. As it was discussed before, OPA2 method
failed to yield reasonably good initial estimates, and a subsequent ALS method could
not conduct the solution towards the correct one (independently if it is ALS1 or ALS2).
In these cases, the experimental decays found were too correlated, and, therefore P was
nearly singular. This caused a significant variance inflation when P is inverted -Eq. (8)-
carrying out a significant error in D (NMR spectra). On the other hand, ALS2 methods
do not result in pure exponential decays (P is unconstrained at the end of one iteration),
so both T2a and T2b obtained do not correspond with the final result. Therefore, only
OPA1-ALS1 is competitive with non-linear, performing the former worse than the
latter.
The only inconvenience of LINOL (compared to ALS-based methods) is the
computation time. Mean computation times for the four temperatures were 10 s (OPA1-
ALS1) 2 s (OPA1-ALS2) 1 s (OPA2-ALS1) 1 s (OPA2-ALS2) and 98 s (LINOL).
OPA-based algorithms terminated after two iterations.
One can think that operating the ALS-based algorithms with constrained D (applying
non-negativity to the solutions) can conduct the ALS algorithm to a better solution.
Table A2.4 presents the SSR obtained with the four algorithms. The quality of the
results was indeed increased in OPA2-ALS1 algorithm, but however the results were in
general still the worse: applying the non-negativity constraint could not solve the
problem of bad initial estimates. The rest of the four algorithms yielded a higher SSR
when constraints are applied. As commented, the more the constrained the solution is,
the higher the SSR will be. This does not happen in OPA2-ALS1, since the (bad) initial
estimates are improved by means of imposing constraints in D. In the case of OPA2-
Appendix 2
157
Figure A2.1. Results of a two-component decomposition after applying the five algorithms described in this article to the NMR experiments at T = 505 K. Parts a-e represent NMR peak profiles, whereas parts f-j depict time decays. Algorithms: a,f OPA1-ALS1; b,g OPA1-ALS2; c,h OPA2-ALS1; d,i OPA2-ALS2; e,j non-linear algorithm.
Appendix 2
158
ALS2 this last effect is less noticeable since the SSR are computed in the final result
(with constrained D), which certainly increases the residuals. Computation times
increased in relation with unconstrained D: 24 s (OPA1-ALS1) 8 s (OPA2-ALS1) 2 s
(OPA2-ALS2) and 10 s (OPA2-ALS2).
Table A2.4. SSR values obtained with the four ALS algorithms. Non-negativity constraint was applied to D.
Temperature OPA1-ALS1
OPA1-ALS2
OPA2-ALS1
OPA2-ALS2
1 1.91 0.47 7.55 1.06
2 2.22 0.17 1.09 0.18
3 3.39 0.32 10.75 5.60
4 0.86 0.16 2.60 1.50
4. Conclusions
MCR methods are a family of techniques aimed to separate the different sources of
variance of bilinear data. This can be used to separate the (partially overlapped) signals
obtained with second-order instruments. Either soft- or hard-modelling can be applied to
MCR methods. When hard-modelling is applied, a mathematical model is applied inside
the MCR algorithm to fit the data (as an extra constrain). The application of hard-
modelling opens the possibility to re-design the MCR algorithm with a different
architecture. Instead of fitting the model inside the MCR algorithm, the MCR algorithm
is included inside a non-linear fitting method. The non-linear method finds the
parameters of the mathematical model, and a linear regression (half of a loop of the
MCR algorithm) is applied each time the function is evaluated.
The method has been tested with solid-state Hahn-Echo NMR data from polymer
networks. This kind of experiments produces two-way data, which can be treated as
bilinear. The goal of the experiment was to separate the contributions of different parts
of a polymer network. In this particular example, the non-linear method performed
Appendix 2
159
better than the MCR algorithm. The method might be applied to other examples of hard-
modelling of bilinear data.
Appendix 2
160
References
[1] Booksh, K. S.; Kowalski, B. R. K. Anal. Chem. 1994, 66, 782.
[2] Cuesta-Sánchez, F.; Toft, J.; van den Bogaert, B.; Massart, D. L. Anal. Chem. 1996, 68, 79.
[3] de Juan, A.; Tauler, R. Anal. Chim. Acta 2003, 500, 195.
[4] Faber, N. M.; Bro, R.; Hopke, P. K. Chemom. Intel. Lab. Syst. 2003, 65, 119.
[5] Erickson, B. C.; Pell, R. J.; Kowalski, B. R. Talanta 1991, 38, 1459.
[6] de Juan, A.; Maeder, M.; Martínez, M.; Tauler, R. Chemom. Intel. Lab. Syst. 2000, 54, 123.
[7] Press, W. H.; Teukolsky, S. A.; Vetterling, W. T.; Flannery, B. P. Numerical Recipes in C, Second Edition, Cambridge University Press 1992.
Appendix 3
161
Appendix 3
Calculation of the (mean) number of monomer units between two cross-linked
junctions
To better interpret the NMR spectral data and the T2H-relaxation measurements we
propose a simulation, based on the assumption of a total (100%) conversion of the
network (full cross-linking). Starting with the relative weight of the monomer units and
their molecular weights, mol percents of these monomers in the unsaturated polyester
network (UP) were obtained (see Table A3.1).
Figure A3.1. Chemical structure of UP network.
Due to the excess of styrene to MA one could expect, in the ideal cross-link reaction,
that all the double bonds of MA have reacted to form the polyester/PS chemical cross-
linked network (Figure A3.1).
Appendix 3
162
Table A3.1. mol % of monomer in the (UP) network
Monomer unit OPA PG MA PS EG Sum
MG of units 148 76 98 104 62
Weight-fraction of 5 1.5 0.5 1.0 2 5
Weight-% 19.2 6.4 12.8 36 25.6 100
n mol in 100 gr net 0.130 0.084 0.131 0.346 0.413
From this assumption, the estimated mean number of styrene units per MA units can be
estimated, as well as the mean number of polyether length per MA-unit:
- PS/MA=0.346/0.131 = 2.64. (1)
- (PG+EG) / (MA) = (0.084+0.413) / (0.131) =3.79. (2)
One may note that (1) can also correspond to the number of PS dangling ends.
Furthermore, the mean number of monomer units in the polyester between two MA-
cross-links can also be derived:
(OPA+PG+MA+EG) / (MA) = (0.130+0.084+0.131+0.413) / (0.131) = 5.79 (3)
No crystallinity in this chemical network should be expected. The mean lengths of the
cross-links are too short. The PS cross-linked junctions correspond to (2*2.64) = 5.28
monomer units and the mean length between the cross points of the polyester backbone
corresponds to 5.79 (3) monomer units as mentioned above. No long linear chains that
can crystallize are present. The extra methyl group of the PG-units is also in favor of
the amorphous phase.
Above Tg (e.g. at 450 K) we expect a comparable molecular mobility for the polyester
backbone and for the PS cross-links, but a different mobility (faster) for the dangling
ends.
The proton spectra above Tg most probably will show the mobile dangling ends,
composed of aromatic signals of PS and OPA around 7 ppm, of ending polyether signals
Appendix 3
163
around 4 ppm and between 1 and 2 ppm the Me-signals from PG together with CH2-
signals from PS-ends.
Table A3.2. Estimation of proton mol % of full cross-linked polyester / polystyrene network
Type of
protons
shifts
in ppm structure mols * n protons
n mol
protons
proton
mol %
Aromatic +/- 7 PS + OPA .346*5+.130*4 2.25 38.5
Glycolic +/- 4 PG + EG .084*3+.413*4 1.90 32.5
Aliphatic 1 to 2 PG + PS + MA .084*3+.346*3+.131*3 1.69 28.9
Acknowledgments
165
Acknowledgements
With pleasure, I come to the acknowledgements.
This thesis carries only my name as author but is the achievement of my own effort
together with several contributions. I am deeply indebted to many people who shared
my Dutch experience so far and made this PhD successful.
I would like to dedicate the first tribute to my promotor Prof.dr. J. Gelan: without your
help, availability, and guidance, this PhD could not have been completed. Your trust
together with the many discussions we had in Hasselt, strengthened my determination: I
owe you a big thank you.
I also would like to express my sincere appreciation to my copromotor Prof.dr. P.J.
Lemstra. You gave me the opportunity to join your research group SKT to carry out my
PhD work. Thank you for your frank approach, and for your always smart and
appropriate observations.
I wish to extend my thanks to the members of the Committee for reading my thesis and
providing constructive comments. My gratitude goes especially to Prof. J. Amouroux:
you accepted me in your prestigious department at Ecole de Chimie Paris and you
considerably helped to shape my career. Your devotion to science and students made
you a reference to many. Also, special thanks to Prof. Hans Geerlings: Thank you
Professor for your availability to discuss, and willingness to help and share your
knowledge. Last but not least I would like to thank Prof.dr. Adriaensens, Dr.ir.
Goossens, and Dr.ir. Magusin for sharing their advice and comments.
I am grateful to all the members of the SKT department. From all of you, I had
something to learn. Thank you for creating such an inspiring working atmosphere, and
for being always so supportive and helpful. In particular, many thanks to Otto van
Asselen for helping me with the FTIR experiment, also for his special kindness and
availability. A big thank to Jules H & K, Martijn, Lijing, Thierry, Joost, Blanca, Casper,
a special thanks to Elly and Tran for helping arranging the defense; also thanks to Bob,
Ineke, Said, and finally to my dearest Cees and Irina. I am also extremely thankful to
Acknowledgments
166
Jacques Schievink from VSSD, the Lucky Luke of the printer: without your flexibility I
would have never made it on time.
Like all experimental projects, this project required much cooperation, with different
Universities and Institutes. Thank you Brahim from Molecular Heterogeneous Catalysis
group in TUe (SKA): your help with the NMR experiments was very much appreciated.
My gratitude goes also to all former and actual members of the Polymer Analysis Group
in Amsterdam (UvA). I would like to particularly thank Prof. Schoenmakers, Wijbren
for your constant and invaluable help in the laboratory, Wim, Philippo, Mauro and also
you, Gabriel Vivo, for your amazing skills and for helping me with the data treatment
part. Next, I wish to record my very sincere and grateful thanks to Prof. B. Jerome from
the FNWI-UvA: you have been so supportive and have always provided very strong
advice. A Woman in Science, and an example to me! I would also like to express my
acknowledgements to the DPI (Dutch Polymer Institute) for financing this PhD project;
to Y. Krager and P. Steeman from DSM, for providing me with samples and for helping
me with the dielectric spectroscopy experiments (DES). Furthermore, thank you to J.
Bloomberg from Shell, for making available the cryogenic grinder.
I experienced the same friendly atmosphere during my work at the Chemistry
Department of Hasselt University (Diepenbeek campus): I offer my most sincere thanks
to the members of the NMR team for their help and guidance.
I am also extremely grateful to Shell for given me the opportunity to finish my thesis
while working for the Company. I am particularly in debt to Rini Reynhout: thank you
Rini for your endless support and precious advice, I will carry them with me forever;
also many thanks to Gerard de Nazelle: your passionate dedication to science and your
determination served to strengthen my own determination to finish. Your support was
crucial in completing this manuscript.
I would also like to express appreciation to Sander van Bavel: thank you Sander for
being such a good friend and for sharing your own PhD experience to help me progress.
I cannot forget Hilbrand Klaver: thanks especially for your patience and bravery to deal
with my high level of stress while sharing the same office. I cannot forget to thank Wim
Wieldraaijer, Frank Niele, for being so inspiring; Michael, Heiko, Alfred, Zakaria,
Patrice, Wouter, Minke, Jan, Joost, Graham and Shanjoy for the many advices. Thank
Acknowledgments
167
you as well to Hans Gosselink, Herman Kuipers, Herman van Wechem and Hans
Stapersma for their flexibility and support.
This thesis meant also for me the opportunity to discover The Netherlands and enjoy the
Dutch way of living (how many bicycles did I buy and lose in Amsterdam?…). I am so
thankful to all of you, who have been part of my Dutch life, and have filled it with so
many discoveries and adventures, have made it full of great fun and so rich. The list is
very long but I cannot forget Jasper, Robin, Manu, Elio, Luca, Fatima, Javier,
Alexander, Caroline and Stephanie (my Charlie’s angels), Alessandra and Luc, Viney,
Nawar (Noa), Marina, Joost, Marcelino Dorothee, Erika, Estella, Cedric and Karima,
Philippe, Lorn and Alain, Nizar, Karim, Chafik, Kosta, Hugo, Leena, Maikel, Sylvia,
Alex, Lokash, Sebastien, Pilar, Sander and my dear Hawaiian friend, Noah Johnson.
I am particularly grateful to Herman, Suzanne, Maaike, Marie Roelofsen and Bastiaan
who always make me feel at home despite the cultural differences and the Dutch
weather; thanks to Mehdi: “ton soutien et ta patience m’ ont aidée dans des moments
difficiles”; a special thank you as well to Basilius: My dear friend, thank you for the
fantastic years we had, and all your support! I will never forget ‘’The Beach’’. Thanks
to my friend Susan Behr: your advices were very precious and helped me progress. And
last but not least, thank you to my two paranimfen, Christien Oele and Valentine
Kreykamp. Thank you ladies it is such an honor to have you as my paranimfen, thanks
for being such great friends.
And further South, there you were all my faithful friends in France and in Algeria,
always close to my heart. Again the list is too long but special thanks to Sabry: c’ est toi
qui m’a fait découvrir ce pays et qui m’a aidé à m’y installer. Je n’oublierai jamais! Un
immense merci à oncle Rabah et ma tante Claude, ainsi qu’à Cherifa et Moh, sans
oublier mes chères et tendres tantes Zina et Louisa, mon oncle Mahdi et Danielle,
‘’tonton’’ Youcef, Lilia, Jelel, Selim, Karim, Réda et Mehdi, Setti, Akli, Baya, Tarik,
Mina, Fayna, Hayet, Hocine, Selma, Amar, Bachira, Leila, Hadjira, Nadra, Nezim, Ines,
Meriem K.
This thesis with the accompanying hectic work is now coming to an end: it has been a
large part of my life, during which I have always hoped to one day complete this
manuscript. Yet, three of my closest friends have shown me a much deeper meaning of
Acknowledgments
168
hope, something which echoes Vaclav Havel’s words: “hope is not the conviction that
something will turn out well…” [my thesis for instance…] “… but the certainty that
something makes sense, regardless of how it turns out”. Laurus, Manola, Farah, my
dearest friends: it is your lives, your examples beyond your deaths, which help me make
sense of our world every day – and make me hold on firmly to hope. I offer you this
thesis, with all my hope and efforts.
And finally, last but by every means first, how will I ever be able to tell you all my
thanks, to you my parents, and to you, my brother. You have tirelessly provided me with
the moral support, your encouragement and understanding and above all, your
unconditional love. Un spécial merci à Maman: cette thèse, je te la dois! Merci pour
avoir toujours cru en moi!
Anybody who has been seriously engaged in scientific work of any kind
realizes that over the entrance to the gates of the temple of science are written the
words: ‘Ye must have faith’.
Max Planck.
Curriculum Vitae
169
Curriculum Vitae
The author of this thesis Maya Ziari was born in Algiers, Algeria. After obtaining her
‘baccalaureat’ in Mathematics and Physics in Nice (France), she completed a bachelor
degree (B.Sc) in Physical Chemistry (University Paris V), followed by a master degree
(M.Sc) in Chemical Engineering in 1998 (University Paris VI-ENSCP) under the
supervision of Prof. J. Amouroux within the Plasma Process department. During her
master, she carried out a traineeship at the Institute for Fundamental Electronics in
Orsay (University Paris XI).
At the beginning of 2000 she started working as a Research Associate at the University
of Amsterdam (UvA) in The Netherlands, on a project granted by Dow Chemicals to
investigate Ultra-High-Molecular-Mass (UHMM) polymers using field flow
fractionation (FFF) and light scattering techniques (LALS/MALS).
In 2002, she started a PhD at Eindhoven University of Technology (TU/e) in the
Polymer Technology Department under the supervision of Prof.dr. P. J. Lemstra and
Prof.dr. J. Gelan. During her PhD study, the author also obtained the diploma of the
‘Register Polymerkundige (RPK)’ organized by the ‘National Dutch Graduate School of
Polymer Science and Technology’ (PTN, Polymer Technologie Nederland).
In 2006, she joined in Amsterdam the Department of Innovation and Research (GSIR)
in Shell Global Solutions International BV, as a Research Associate on a contractor
basis. In 2009, she joined Shell (NAM) on a permanent contract and was transferred to a
new role as Appraisal Engineer in the Department of Exploration and Production.