Borosilicones and viscoelastic silicone rubbers: network liquids and network solids

Louis A. Bloomfield∗ Department of Physics, University of Virginia, Charlottesville, VA 22904 (Dated: January 30, 2018) Borosilicones such as bouncing putty (e.g., Silly Putty) have been known for more than 70 years, but the origins of their peculiar behaviors have remained a mystery. In this work, experiment and theory are used to show that borosilicones are network liquids—dynamic macromolecules that appear elastic on short timescales but exhibit flow on longer timescales. Each borosilicone is a vast covalent network of silicone polymer chains joined by trifunctional boron crosslinks. At any instant, the borosilicone is a highly-crosslinked elastic material. Because the boron crosslinks are temporary, however, the network evolves with time and the borosilicone exhibits liquid behavior. A simple borosilicone, with chains coupled only by temporary crosslinks, exemplifies a classic transient network model and behaves as a simple (Lodge) elastic fluid. Its measured moduli and viscosities fit those predicted by both the transient network model and the Maxwell viscoelastic model: an elastic spring in series with a viscous dashpot. Those models lead to the same integer-order linear differential equations, which predict the exponential relaxation processes that are observed. When some of the chains in a borosilicone are permanently crosslinked, however, the borosilicone no longer behaves as a simple elastic fluid. Though still a network liquid, this non-simple borosilicone exhibits slower-than-exponential relaxation processes that cannot arise from the integer-order linear differential equations representing finite arrangements of integer-order viscoelastic element (dashpots and springs). Infinite arrangements have no such limitations and the non-simple borosilicone can be modeled as a random assembly of infinitesimal viscoelastic bodies. Viscous bodies representing transient couplings occupy volume fraction β and elastic bodies representing non-transient couplings occupy volume fraction 1 − β. An analytic study of this random assembly and a computational study of the analogous random network find behavior that is intermediate between viscous and elastic. Whereas a viscous material obeys differential equations of order 1 and an elastic material obeys differential equations of order 0, the random assembly obeys differential equations of fractional order β. The random assembly acts as a fractional-order viscoelastic element known as a spring-pot. To eliminate the spring-pot’s unphysical divergence as ω → ∞, a spring is placed in series with it. That pairing is the Fractional Maxwell viscoelastic model, the predictions of which agree well with the measured moduli and viscosities of non-simple borosilicones. Among those predictions are time-dependent viscosities that grow without limit and Mittag-Leffler-function relaxation processes. When its concentration of permanent crosslinks exceeds the gelation threshold, a borosilicone becomes a viscoelastic silicone rubber (VSR). With a permanent network that spans the material coupled to a temporary network that also spans the material, the VSR is a network liquid piggy- backing on a network solid. As a network solid, the VSR has an equilibrium shape to which it returns in the absence of external influences. As a network liquid with transient and non-transient couplings, the VSR has viscoelastic dynamics like those of a non-simple borosilicone. The Fractional Zener viscoelastic model, an elastic spring in parallel to the Fractional Maxwell model, accurately predicts the measured moduli of VSRs. The temporary nature of boron crosslinks is due to exchange reactions in which -OH bearing molecules substitute for another endlessly in their covalent attachments to boron atoms. The en- ergy barrier is small, so the mean lifetime of the temporary crosslinks is only modestly temperature- dependent. Because that mean lifetime gives rise to a borosilicone’s only significant timescale, dom- inating its viscoelastic dynamics, a change in temperature merely shifts the borosilicone’s timescale and the borosilicone is therefore thermo-rheologically simple.

INTRODUCTION tem, and offer unparalleled physical realizations of sev- arXiv:1801.09253v1 [cond-mat.soft] 28 Jan 2018 eral basic viscoelastic models.

Borosilicones are a unique class of viscoelastic liquids While the simplest borosilicones can be described using and solids. Misunderstood and neglected scientifically for the models and mathematics of viscoelasticity texts, in- about seventy years, borosilicones are actually far more cluding differential equations of integer order, more com- interesting than they originally appeared. In their sim- plicated borosilicones require more sophisticated mod- plest form, borosilicones are silicone polymer chains con- els and mathematics. These include differential equa- nected temporarily by boron crosslinks. In addition to tions of non-integer order, random network theory, and having exceptional physical and mechanical properties, fractional viscoelastic model elements. Because the ex- borosilicones exemplify some of the most basic and pow- perimental measurements of borosilicones are best inter- erful theories in rheology, viscoelasticity, and random sys- preted through those theoretical models and calculations, 2 experiment and theory are woven together throughout The transient network (TN) model makes five key this article. assumptions[22]: Borosilicones were first prepared in the 1940s[34] and 1. The strands are Hookean entropy springs their unusual properties were soon recognized with the discovery of silicone bouncing putty (SBP)[47]. A rudi- 2. The strands deform affinely until they release mentary borosilicone, SBP is a liquid that collapses 3. Strands break with a constant probability per unit slowly into a puddle under its own weight yet bounces time, independent of the network deformation beautifully as a ball dropped onto a hard surface. SBP eventually became the key ingredient in Silly Putty, a fa- 4. Strands re-form as fast as they break miliar childrens toy, and is sold in various forms to this 5. Strands re-form in configurations typical of equilib- day. Despite its unusual properties, however, SBP found rium little practical use outside the toy store and has remained mostly a curiosity for more than half a century. If the probability per unit time of a strand breaking Several explanations for SBPs unusual behavior have and reforming is 1/τ, then the probability P (t − t0) of a been proposed over the years[14, 28, 45, 49], with the cur- strand surviving intact from time t0 to time t satisfies the rent favorite being that SBP is composed of long poly- equation mer molecules that are entangled and linked by hydrogen d 1 bonds[9]. On short timescales, the hydrogen bonds and P (t − t0) = − P (t − t0). (1) dt τ entanglements prevent flow and the SBP behaves as an elastic solid. On longer timescales, the hydrogen bonds Since P (0) = 1, the solution to Eq. (1) is 0 can break, the molecules can disentangle, and the SBP P (t − t0) = e−(t−t )/τ . (2) can flow. SBP has been the textbook example of a ther- moplastic polymer in the rubbery flow regime[41], a cat- The time constant τ is thus the mean lifetime of the egory it shares with the polymers found in chewing gum. strands. ↔ This paper will show that SBP is actually a Lodge We can calculate the contribution dσ t0 to the mate- ↔ Elastic Fluid, a viscoelastic fluid defined by a transient rial’s stress tensor σ(t) due to a strand that re-formed at network theory first proposed by Green and Tobolsky in time t0 as the product of four quantities: the probability 1946[15] and extended by Lodge in 1956[24]. In their dt0/τ that the strand formed during the interval between original transient network model, Green and Tobolosky t0 and t0 + dt0, the probability P (t − t0) that this strand explored the behavior of elastic strands networked to- survived until time t, the material’s modulus G, and the ↔−1 gether by chemical crosslinks that break and reform at Finger tensor C (t, t0) characterizing the deformation a steady rate. Their work served to explain the slow that occurred between time t0 and time t stress relaxation observed in polysulfide rubbers at ele- 0 ↔−1 ↔ dt −(t−t0)/τ 0 vated temperatures.[42] dσ 0 = e GC (t, t ). (3) t τ Although Green and Tobolsky’s transient network model assumed chemical crosslinks, it found much wider From rubber elasticity theory, the modulus G = νkT , application when Rouse[35], Zimm[50], and Lodge[24] where ν is the number of strands per unit volume. Inte- 0 extended it to molten polymers and polymer solutions. grating Eq. (3) over all past times t gives the constitutive Though not crosslinked in a chemical sense, the strands equation for the transient network model, also known as in molten polymers and polymer solutions interact with the Lodge equation: one another in ways that resemble transient crosslinks, Z t ↔−1 ↔ 1 −(t−t0)/τ 0 0 notably through entanglements. Fluids exemplifying the σ(t) = G e C (t, t )dt . (4) −∞ τ transient network model are known as Lodge Elastic Flu- ids. In general, a viscoelastic material’s stress-strain rela- tionship can be written[39]

Z t   ↔−1 ↔ ∂ 0 0 0 LODGE ELASTIC FLUID σ(t) = 0 G(t − t ) C (t, t )dt , (5) −∞ ∂t Since it will be shown that silicone bouncing putty is where G(t − t0) is the stress relaxation modulus. For a Lodge Elastic Fluid (LEF), a brief review of the tran- the Lodge Elastic Fluid described by Eq. (4), the stress 0 sient network model is appropriate. That model builds relaxation modulus Glef(t − t ) is on rubber elasticity theory, particularly the kinetic the- 0 G (t − t0) = Ge−(t−t )/τ . (6) ory of elasticity[40], which finds that polymer strands of lef sufficient length and flexibility have a Gaussian distribu- Thus, the stress in an LEF produced by a step in strain at tion of end-to-end separations and behave as Hookean time t0 = 0 decays away exponentially with characteristic entropy springs. time τ, the mean lifetime of strands in the LEF. 3

SIMPLE BOROSILICONES To produce a saturated borosilicone, 8.13g TMB (Alfa- Aesar B20215) were added to 100.00g water-free An- There are a number techniques for incorporating disil OH 40 and the mixture was vacuum dried through boron in polyorganosiloxanes[30] and consequently for two liquid-nitrogen-cooled Pyrex cold traps. Over a 30 producing silicone bouncing putties[7, 11, 27, 29]. In minute period, 6.68g methanol were collect in those traps, the present work, simple borosilicones were prepared 89 wt% of the methanol expected if the condensation re- by reacting a silanol-terminated polydimethylsiloxane action went to completion. By that time, the borosilicone fluid OH-PDMS-OH with the liquid boron compound had become a brittle, translucent foam that made remov- ing additional methanol difficult. The borosilicone was trimethyl borate B(OCH3)3. Rochow used a similar tech- nique in 1940 to produce the first borosilicones[34], but crushed and further vacuum dried, but the the residual did not immediately realize what had occurred. Re- methanol was not collected. actions of silanol groups with alkoxy groups on boron This simple borosilicone, designated SB40-8.13, is brit- were thought to have relatively low yields[1, 30], how- tle, granular, and easily mistaken for a solid. It is actu- ever, Boot successfully produced SBPs in 1965 using such ally an extraordinarily viscous liquid. When compressed reactions[7] and it works well in practice. in an arbor or hydraulic press, its granules slowly consoli- The reaction between a silanol group (OH-Si) and a date into a transparent mass. Moreover, SB40-8.13 flows under its own weight, but at such an imperceptible rate methoxy group on boron (RO-CH3) is a condensation reaction that days are required to observe any significant change in shape. [N.B. liquid flow is universal among simple

B(OCH3)3 + HO-PDMS-OH borosilicones—despite hundreds of experiments in which l many different STPDMS fluids were combined with vari- B(OCH3)2-O-PDMS-OH + CH3OH ous amounts of TMB and then fully dried, no solid simple borosilicones were ever observed.] and results in a methanol molecule CH3OH. This reaction That SB40-8.13 remains liquid, despite being at or near is reversible, so that alcoholysis accompanies condensa- stoichiometric saturation, defies the predictions of Flory- tion, and it normally reaches an equilibrium in which Stockmayer theory[12, 13, 43]. According to that theory, both reactants and products are present. a linear polymer (e.g. STPDMS) crosslinked by a tri- To drive the reaction toward condensation and borosil- functional branching unit (e.g. boron) reaches the gela- icone formation, the methanol must be removed. Simply tion threshold at only 50% of stoichiometric saturation. allowing the methanol to evaporate is sufficient, but vac- At 100% of stoichiometric saturation, it should be a net- uum drying propels the reaction to completion far more work solid—a gigantic macromolecule in which a single efficiently. The condensation reaction is so rapid, even at covalently-bonded network extends throughout the entire room temperature, that vacuum removal of the methanol material. With nearly all of its PDMS chains connected can cause the borosilicone to form in seconds. Vacuum by boron crosslinks, SB40-8.13 should be a network solid drying was used routinely throughout this work. and unable to flow. To produce borosilicones of the highest quality, only The answer to this puzzle lies in the boron crosslinks unblended, single-equilibration silanol-terminated poly- themselves. Boron crosslinks are not permanent, they are dimethylsiloxane (STPDMS) fluids were used in their temporary. Unlike conventional crosslinks, which rarely preparation. Though commercial products rather than break in normal circumstance, boron crosslinks detach pure chemicals, these single-equilibration fluids had rela- and re-attach on timescales measured in seconds or less. tively narrow, single-peaked molecular weight (MW) dis- With its temporary crosslinks, SB40-8.13 is not a net- tributions and minimal impurities. work solid at all, it is a network liquid—a gigantic macro- The concentration of OH groups in each fluid could molecule in which a single covalently-bonded network ex- be estimated from its viscosity and used to determine tends throughout the material, but with covalent bonds the amount of trimethyl borate (TMB) needed to achieve that detach and re-attach frequently so that the network stoichiometric saturation. When 3 mol of STPDMS react can evolve in topology and geometry. with 2 mol of TMB and the methanol is removed, all of The mechanism by which boron crosslinks detach and the silanol groups of the STPDMS fluid (HO-PDMS-OH) re-attached almost certainly involves ligand substitution are replaced by boron crosslinks (B-O-PDMS-O-B) and rather than actual bond breaking. In this substitution stoichiometric saturation is achieved. process, a =BO- moiety on a boron crosslink aligns with Andisil OH 40 (AB Specialty Silicones) is a low-MW an -OH group on another molecule and, facilitated by STPDMS fluid which nominally contains 3.5 wt% OH their lone electron pairs, the two oxygen atoms trade groups. The product lot used in this work was certified bonding partners (Fig. 1). The boron crosslink is left as containing 3.99 wt% OH, a value confirmed locally with a new ligand. That substitution process is similar using Karl-Fischer titration[5]. Adding 8.13 wt% TMB to the four-center mechanism proposed for the rapid re- to that fluid should result in stoichiometric saturation. distribution of alkoxy groups on boron observed in boron 4

H H 0 PDMS O OR PDMS O OR 10 BO BO 10 -1 SB40-8.13 G(t) PDMS O PDMS PDMS O PDMS (MPa) G = 1.12 MPa -2 G (t) 10 τ = 16.7 s LEF G(t)

FIG. 1. A proposed mechanism for the detachment and re- 10 -3 attachment of boron crosslinks. When an -OH group on a 10 -2 10 -1 10 0 10 1 10 2 molecule aligns with a =BO- crosslink, the lone electron pairs Time (s) on the two oxygen atoms facilitate a rearrangement of bonds. The previously unattached molecule crosslinks to the boron FIG. 2. Stress relaxation modulus G(t) of SB40-8.13, fit by and a previously crosslinked ligand is released with a new - the stress relaxation modulus Glef(t) of an LEF. Modulus G OH group. In this substitution, HO-R can be either a small and characteristic time τ are the fit’s two parameters. The molecule (e.g., water, an alcohol, or a carboxylic acid) or a SB40-8.13 was in equilibrium with laboratory air. silanol-terminated PDMS chain.

where λ is the final height divided by initial height. Com- esters[17]. bining Eqs. (8) and (7) gives As evidence for such ligand substitution, it will be Z 0     shown that the rate at which boron crosslinks detach ∂ 0 2 1 0 σc(t) = 0 G(t − t ) λ − dt and re-attach is extremely sensitive to the concentration −∞ ∂t λ of certain -OH groups, particularly -COOH groups, in  1  a borosilicone. Adding just 1 ppm of iso-stearic acid = − λ2 − G(t), (9) λ (C17H35COOH) to a borosilicone significantly increases this rate. That -COOH groups boost the substitution so that rate more than other -OH groups probably stems from  λ  the second oxygen offering more possibilities for proper G(t) = −σ (t) . (10) c 1 − λ3 alignment and partner exchange. It will also be shown that the activation energy for boron crosslink detachment and re-attach is far too small to involve the breaking a Figure 2 shows SB40-8.13’s stress relaxation modulus covalent bond. G(t), obtained using the sudden compression technique. Also shown is a fit to those data by Glef(t), the stress relaxation modulus of an LEF from Eq. (6). The quality Stress Relaxation Modulus of the fit supports the hypothesis that SB40-8.13 is an LEF. It also provides values for the modulus G and the characteristic time τ. Repeated measurement yielded an A borosilicone’s stress relaxation modulus G(t) is mea- experimental value of G = 1.135 ± 0.048 MPa. Here sured by subjecting a small cylinder of that borosilicone and throughout this work, reported uncertainties are 2 to sudden compression. The 9.53mm-diameter cylinder standard deviations (95% confidence). is compressed from 6.35mm tall to 4.98mm tall in 10ms For comparison, Andisil OH 40 was also cured with a and held at that height while a load cell records the com- conventional trifunctional crosslinker, methyltriethoxysi- pressive force as a function of time. Dividing the com- lane (MTEOS). Using the technique of Mark[25], 1.0 wt% pressive force by the cylinder’s cross sectional area yields tin 2-ethylhexanoate catalyst were mixed into Andsil OH the compressive stress σ (t), from which the stress re- c 40, followed by 10.66 wt% MTEOS (stoichiometric satu- laxation modulus G(t) can be calculated. Effects due ration), and the mixture was cured in vacuum at 20 ◦C to cylinder barreling, Coulomb friction[46], and load-cell for 10 days. deformation are neglected. The cured silicone rubber was studied in the same com- When a cylinder’s flat surfaces are perpendicular to pression apparatus used for the borosilicones. It was axis 2 and compressed along axis 1, its total compressive found to have modulus G = 0.907 ± 0.030 MPa and no stress σ (t) = σ (t) − σ (t).[39] Using Eq. (5), c 11 22 significant stress relaxation. That modulus is slightly less Z t  ∂  than the modulus of SB40-8.13, evidence that SB40-8.13 σ (t) = G(t − t0) C−1(t, t0) − C−1(t, t0)
dt0. is a covalently-bonded network liquid. Rubber elasticity c ∂t0 11 22 −∞ theory predicts G ≈ 3.0 MPa. (7) After vacuum drying, SB40-8.13 has a characteristic For a compression step at t0 = 0, the Finger tensor ele- time τ of about 40 s. τ is extremely sensitive to certain ments are -OH groups, however, so it decreases as the borosilicone 1 absorbs moisture from the air. When SB40-8.13 was al- C−1(t, t0) − C−1(t, t0) = λ2 − [t0 < 0] 11 22 λ lowed to equilibrate with laboratory air (20 ◦C, 50 % = 0, [t0 ≥ 0] (8) relative humidity), its τ decreased to about 17 s. 5

Shear Viscosity 20 η(t) η (t) The shear viscosity η(t) of a viscoelastic fluid is the LEF 10 γ· = 0.000003 s -1 ratio of shear stress σ (t) to shear strain rateγ ˙ , (MPa·s) s SB40-8.13 τ = 48 s η(t) ηLEF = 17.8 MPa·s σs(t) η(t) = , (11) 0 γ˙ 20 for constant-rate shear strain that begins at time t = 0. η(t) is time-dependent because a viscoelastic fluid’s shear 10 γ· = 0.00003 s -1 stress σs(t) depends on its shear strain at all earlier times (MPa·s) τ = 30 s η(t) and thus on t, the time since the start of shearing. ηLEF = 17.4 MPa·s A borosilicone’s shear viscosity η(t) is measured by 0 subjecting a 33mm-diameter by 4.0mm-thick disk to 20 steady shear starting at time t = 0, while a load cell records the shear force as a function of time. Dividing 10 γ· = 0.0003 s -1

that shear force by the disk’s cross sectional area and (MPa·s) τ = 29 s thickness gives the shear stress σs(t), which is then di- η(t) ηLEF = 17.2 MPa·s vided by the shear strain rateγ ˙ to obtain the shear vis- 0 cosity η(t). Load cell deformation during the measure- 20 ments is often significant enough to require compensation in order to maintain shear rates that are constant and ac- curate. 10 γ· = 0.003 s -1 (MPa·s) τ = 28 s The time-dependent shear viscosity ηlef(t) of an LEF η(t) η = 16.7 MPa·s can be calculated for comparison, beginning with a gen- LEF 0 eral viscoelastic fluid disk. When that disk’s flat surfaces 0 20 40 60 80 100 are perpendicular to axis 2 and sheared along axis 1, its Time (s) shear stress σs(t) = σ21(t)[39] and Eq. (5) becomes FIG. 3. Shear viscosity η(t) of SB40-8.13 at four different Z t   ∂ 0 −1 0 0 shear ratesγ ˙ . Each measurement is fit by the shear viscosity σs(t) = 0 G(t − t ) C21 (t, t )dt . (12) −∞ ∂t ηlef(t) of an LEF. Limiting viscosity ηlef and characteristic time τ are the fit’s two parameters. The SB40-8.13 was drier For shear strain at constant rateγ ˙ beginning at t0 = 0, than laboratory air.

−1 0 0 C21 (t, t ) =γt ˙ [t < 0] =γ ˙ (t − t0). [t0 ≥ 0] (13) The shear viscosity of an LEF thus starts at zero and approaches its long-time limit η = Gτ exponentially Combining Eqs. (12) and (13) gives lef with characteristic time τ. Since ηlef is independent of Z t  ∂  the shear rateγ ˙ , an LEF exhibits neither shear thinning σ (t) = G(t − t0) γtdt˙ 0 s 0 nor shear thickening. −∞ ∂t Measurements of the shear viscosity of SB40-8.13 at 4 Z t  ∂  0 0 0 different shear ratesγ ˙ are shown in Fig. 3, along with fits − 0 G(t − t ) γt˙ dt 0 ∂t by ηlef(t). The fits are excellent, further evidence that Z t SB40-8.13 is an LEF, and they provide values for the =γ ˙ G(t0)dt0, (14) 0 long-time shear viscosity ηlef and the characteristic time τ. η decreases by only 6% as the shear rate increases where it has been assumed that tG(t) → 0 as t → ∞. lef by three orders of magnitude, nearly achieving the LEF Combining Eqs. (11) and (14) gives prediction of no shear thinning at all. Z t The characteristic time τ decreases with increasing 0 0 η(t) = G(t )dt . (15) shear rate, particularly betweenγ ˙ = 0.000003s−1 and 0 γ˙ = 0.00003s−1. This effect is largely an instrumental ar- For an LEF, Glef(t) is given by Eq. (6) and the time- tifact because, at the lowest shearing rates, the shearing dependent shear viscosity ηlef(t) is motion is so slow that elastic deformation of the “rigid” Z t instrument becomes significant. Although compensation −t0/τ 0 ηlef(t) = G e dt is made for the increasing load cell deformation early in 0 each measurement, some instrumental deformations re- = Gτ(1 − e−t/τ ). (16) main uncompensated and affect τ in the slowest shear 6 measurements. where it has been assumed that e2t ˙ − e−t˙
G(t) → 0 The modulus G of SB40-8.13 obtained by dividing as t → ∞. Combining Eqs. (20) and (17) gives η by τ for many shear viscosity measurements is lef Z t Z t G = 0.584 ± 0.141 MPa. This G is somewhat less than 0 2t ˙ 0 0 0 −t˙ 0 0 ηe(t) = 2 G(t )e dt + G(t )e dt . (21) that obtained using the sudden compression technique. 0 0 This difference is probably due to sample edge effects, barreling, and friction[46], and to imperfect equipment. For an LEF, Glef(t) is given by Eq. (6) and the time- Improving absolute accuracy will be a focus of future dependent elongation viscosity ηe,lef(t) is studies. Z t Z t −t/τ 2t ˙ 0 0 −t/τ −t˙ 0 0 ηe,lef(t) = 2G e e dt + G e e dt 0 0 2Gτ Elongation Viscosity = (1 − e−(1−2τ ˙ )t/τ ) 1 − 2˙τ Gτ The elongation viscosity ηe(t) of a viscoelastic fluid is + (1 − e−(1+τ ˙ )t/τ ). (22) the ratio of elongation stress σe(t) to elongation strain 1 +τ ˙ rate ˙, When 2˙τ < 1, ηe,lef(t) asymptotically approaches a σe(t) finite limit ηe,lef at long times ηe(t) = , (17) ˙ 1 η = 3Gτ . (23) for constant-rate elongation strain that began at time t = e,lef (1 − 2˙τ)(1 +τ ˙ ) 0. ηe(t) is time-dependent because a viscoelastic fluid’s elongation stress σe depends on its elongation strain at When 2˙τ ≥ 1, however, ηe,lef(t) increases without limit all earlier times and thus on t, the time since the start of and does so exponentially in time. That effect is known elongation. as strain hardening[22] and leads inevitably to a broken A borosilicone’s elongation viscosity ηe(t) is measured beam if that beam is subjected to constant-rate elonga- by stretching a rectangular beam so that its length in- tion at ˙ ≥ 1/2τ. creases exponentially in time, as required for constant- Measurements of the elongation viscosity of SB40-8.13 rate elongation. The beam is initially 6.54 mm x 6.54 at 4 different elongation rates ˙ are shown in Fig. 4, mm x 14.22 mm long (L0), and mounted in a stretching along with fits by ηe,lef(t). The fits provide values for apparatus with the help of dog-bone ends. Starting at the modulus G and the characteristic time τ, from which time t = 0, this beam is stretched exponentially so that SB40-8.13’s limiting elongation viscosity ηe,lef has been t˙ its length at t > 0 is L(t) = L0e . The exponential calculated using Eq. (23). That elongation viscosity de- stretching continues for a specified time unless the beam creases slightly with increasing elongation rate. The val- breaks or the apparatus reaches its mechanical limits. ues obtained for modulus G range too widely to provide The time-dependent elongation viscosity ηe,lef(t) of an a meaningful average and confidence interval. LEF can be calculated for comparison, beginning with a Unreinforced borosilicones have poor tensile strengths general viscoelastic beam. When that beam is elongated and SB40-8.13 beams routinely break at elongation rates −1 along axis 1, its elongation stress σe(t) = σ11 − σ22 and greater than about 0.015 s . It was thus not possible to Eq. (5) becomes exceed the strain hardening threshold 2˙τ ≥ 1, which is ˙ = 0.125 s−1 when τ = 4 s. Adding reinforcing fillers to Z t  ∂  0  −1 0 −1 0  0 this borosilicone would allow measurements in the strain σe = 0 G(t − t ) C11 (t, t ) − C22 (t, t ) dt . −∞ ∂t hardening domain, however, those fillers introduce other (18) important effects that will be discussed below. For constant elongation strain rate ˙, beginning at t0 = 0, −1 0 −1 0 C11 (t, t ) − C22 (t, t ) = Complex Shear Modulus e2t ˙ − e−t˙ [t0 < 0] (19) 2 ˙(t−t0) −˙(t−t0) 0 e − e [t ≥ 0] The complex shear modulus G∗(ω) of a viscoelastic Combining Eqs. (18) and (19) gives material is the ratio of shear stress σs(ω, t) to shear strain (ω, t) for steady oscillatory shear strain (t) =  eiωt, Z 0   0 ∂ 0  2t ˙ −t˙  0 σe(t) = G(t − t ) e − e dt ∂t0 σs(ω, t) −∞ G∗(ω) = . (24) Z t   (ω, t) ∂ 0 h 2 ˙(t−t0) −˙(t−t0)i 0 + 0 G(t − t ) e − e dt 0 ∂t A borosilicone’s complex shear modulus G∗(ω) is mea- t t Z 0 Z 0 sured using linear oscillatory shear strain. An aluminum = 2˙ G(t0)e2t ˙ dt0 + ˙ G(t0)e−t˙ dt0, (20) 0 0 septum is sandwiched between two disks of borosilicone, 7

15 coelastic disk. When surfaces perpendicular to axis 2 are η (t) E sheared along axis 1, shear stress σs(t) is given by Eq. 10 ηE,LEF (t) iωt (12). For steady oscillatory shear strain (t) = 0e , SB40-8.13

0

(t) (MPa∙s) (t) 5 ∙ -1 −1 0 iωt iωt E ϵ = 0.001 s G = 0.85 MPa C (t, t ) =  e −  e . (25) η 21 0 0 ηE,LEF = 11.6 MPa∙s τ = 4.5 s 0 15 Combining Eqs. (12) and (25) gives

Z t   10 ∂ 0  iωt iωt0  0 σs(t) = 0 G(t − t ) 0e − 0e dt −∞ ∂t 5 ∞ (t) (MPa∙s) (t) ∙ -1 Z E ϵ = 0.0025 s G = 0.87 MPa iωt 0 −iωt0 0 η η = 9.4 MPa∙s τ = 3.6 s = iω0e G(t )e dt (26) E,LEF 0 15 where it has been assumed that G(t) → 0 as t → ∞. 10 Combining Eqs. (24) and (26) gives

Z ∞ 5 ∗ 0 −iωt0 0 (t) (MPa∙s) (t) ∙ -1 G E ϵ = 0.005 s = 0.97 MPa G (ω) = iω G(t )e dt . (27) η 0 ηE,LEF = 9.2 MPa∙s τ = 3.1 s 0 15 For an LEF, Glef(t) is given by Eq. (6) and the com- plex shear modulus is 10 Z ∞ ∗ −t0/τ −iωt0 0 5 Glef(ω) = iωG e e dt (t) (MPa∙s) (t) ∙ -1 G E ϵ = 0.01 s = 0.53 MPa 0 η η = 7.7 MPa∙s τ = 4.6 s E,LEF iωτ 0 = G . (28) 0 5 10 15 20 25 1 + iωτ Time (s) A complex modulus can be separated into its real and FIG. 4. Elongation viscosity ηe(t) of SB40-8.13 at four dif- imaginary parts ferent elongation rates ˙. Each measurement is fit by the elongation viscosity ηe,lef(t) of an LEF. Modulus G and char- G∗(ω) = G0(ω) + iG00(ω), (29) acteristic time τ are the fit’s two parameters. Limiting viscos- ity η is calculated from those parameters using Eq. (23). e,lef 0 00 To reduce its brittleness, the SB40-8.13 was moister than lab- where G (ω) is the storage modulus and G (ω) is the loss oratory air. modulus. For an LEF, those two parts are

2 2 0 ω τ Glef(ω) = G 2 2 (30) nominally 20mm dia x 1.4mm thick, and that assembly 1 + ω τ ωτ is sandwiched between two aluminum plates affixed to a 00 Glef(ω) = G 2 2 . (31) load cell. An amplified waveform synthesizer and voice- 1 + ω τ coil linear actuator caused the septum to oscillate along A measurement of the complex shear modulus of SB40- its length relative to the fixed plates. ∗ 8.13 is shown in Fig. 5, along with a fit by Glef(ω). The An optical system measures the oscillatory shear strain fit provides values for the modulus G and the character- while the load cell measures the oscillatory shear force. istic time τ, but G is in arbitrary units due to the lack Dividing the oscillatory shear force by the oscillatory of absolute calibration. shear strain gave the complex shear modulus times a cal- At small ω, the agreement between measurement and ibration constant that depended on the exact geometries theory is excellent and SB40-8.13 is consistent with a of the two sample disks. Lodge Elastic Fluid. At large ω, however, the measured 0 0 For technical reasons, that calibration constant proved loss modulus G (ω) exceeds the predicted Glef(ω), sug- too difficult to control or measure accurately, so complex gesting that the simple borosilicone has fast dynamical modulus values are reported in arbitrary units. At each processes that are not included of the transient network angular frequency ω, five or more oscillatory cycles were model. measured and analyzed separately in order to provide a Instrumental resonances sometimes interfere with mea- value and a 95% confidence interval for G∗(ω). surements at specific large values of ω. In Fig. 5, an ∗ −1 The complex shear modulus Glef(ω) of an LEF can instrumental resonance near ω = 628 s disturbed the be calculated for comparison, starting with a general vis- data for that ω. 8

0 10 G'(ω) TABLE I. Viscosities of Andisil OH 40 borosilicones measured -1 10 GLEF '(ω) using a falling sphere viscometer. For each borosilicone, the (au) SB40-8.13 -2 specified amount of trimethyl borate (TMB) was added to 10 G = 2.48 au G''(ω) Andisil OH 40 and the mixture was vacuum dried.

G*(ω) -3 τ = 14 s 10 G ''(ω) LEF Name TMB (wt%) Viscosity (Pa·s) Character -4 10 OH 40 0.00 3.9 · 10−2 thin liquid 10 -3 10 -2 10 -1 10 0 10 1 10 2 10 3 SB40-4.00 4.00 1.4 · 100 syrup ω (1/s) SB40-4.10 4.10 3.0 · 103 soft gum SB40-4.20 4.20 1.3 · 104 firm gum FIG. 5. Complex shear modulus G∗(ω) = G0(ω) + iG00(ω) of SB40-4.50 4.50 6.5 · 105 gel SB40-8.13, fit by the complex shear modulus G∗ (ω) of an lef SB40-7.13 7.13 2.8 · 107a “solid” LEF. Modulus G and characteristic time τ are the fit’s two parameters. Lacking absolute calibration, G is in arbitrary a Measured using a linear shear viscometer units. The SB40-8.13 was in equilibrium with laboratory air.

2

Maximizing the Modulus 1.5

1

A simple borosilicone need not be at stoichiometric (MPa) G saturation to have a substantial modulus. Because boron 0.5 crosslinks are dynamic, a borosilicone’s modulus depends 0 more on the statistics of its crosslinks than it does on 4 4.5 5 5.5 6 6.5 7 7.5 8 8.5 how close the borosilicone is to stoichiometric saturation. TMB (wt%) Since boron crosslinks that coordinate only one or two FIG. 6. Moduli G of Andisil OH borosilicones measured re- PDMS chains contribute little to the modulus, maximiz- peatedly using sudden compression. For each borosilicone, ing the time-average concentration of boron crosslinks the specified amount of trimethyl borate (TMB) was added coordinating three PDMS chains should maximize the to Andisil OH 40 and the mixture was vacuum dried. modulus. Lacking a detailed study of the exchange reactions oc- curring at equilibrium, the moduli of these network liq- traps and totaled 92.3 wt% of the methanol expected uids had to be maximized by trial and error. Borosili- if the condensation reaction went to completion. The cones with different boron concentrations were prepared brittle borosilicone foam was then crushed and further from the same lot of Andisil OH 40 fluid and their vis- dried to remove residual methanol. It has modulus cosities and moduli measured. G = 1.366 ± 0.140 MPa. Near the gelation threshold, the borosilicones are soft For comparison, a silicone rubber equivalent to SB40- enough that their viscosities can be measured with a 7.13 was prepared using MTEOS and the curing pro- falling sphere viscometer (Table I). For a trifunctional cedure described above. It was found to have modulus crosslinker such as boron, the gelation threshold is 50% G = 0.877 ± 0.020 MPa, significantly less than the mod- of stoichiometric saturation[12, 13]. Since the Andisil OH ulus of its borosilicone equivalent. 40 saturates at 8.13 wt% TMB, its gelation threshold is That the borosilicone modulus peaks below stoichio- 4.06 wt% TMB. metric saturation suggests that dynamic competition be- The borosilicones indeed show a dramatic increase in tween the abundant boron crosslinks at saturation results viscosity as the gelation threshold is exceed. This obser- in too many of those crosslinks being under-coordinated. vation is consistent with the idea that borosilicones are A small reduction in boron crosslink concentration boosts network liquids and that their covalently-bonded molec- their time-average coordination and results in a some- ular networks grow to macroscopic size once the gelation what larger modulus. threshold is exceeded. Additionally, the saturated borosilicone SB40-8.13 be- Further above the gelation threshold, the borosilicones comes slightly cloudy when allowed to equilibrate with are stiff enough that their moduli can be measured using laboratory air. That cloudiness, almost certainly due to the sudden compression technique (Fig. 6). As the boron tiny crystallites of boric acid, is frequently observed in content increases, the modulus increases to a maximum simple borosilicones with excessive boron concentrations. at 7.13 wt% TMB and then decreases until stoichiometric Dynamic competition between boron crosslinks and wa- saturation is reached at 8.13 wt% TMB (SB40-8.13). ter molecules drives a small fraction of the boron out of The 7.13 wt% borosilicone, designated SB40-7.13, was solution as boric acid. prepared by adding 14.26g TMB to 200.000g Andisil A measurement of SB40-7.13’s stress relaxation mod- OH 40. Methanol released during the vacuum drying ulus G(t) is shown in Fig. 7, along with a fit by Glef(t). of SB40-7.13 was collected in liquid-nitrogen-cooled cold Many measurements yielded G = 1.366±0.140 MPa, the 9

0 0 10 10 G'(ω) SB40-7.13 -1 G '(ω)

-1 (au) 10 LEF 10 G(t) SB40-7.13 (MPa) G = 1.37 MPa 10 -2 G = 1.95 au -2 τ = 15.2 s GLEF (t)

G(t) G''(ω) 10 G*(ω) τ = 8.6 s 10 -3 GLEF ''(ω) 10 -3 10 -4 10 -2 10 -1 10 0 10 1 10 2 10 -3 10 -2 10 -1 10 0 10 1 10 2 10 3 Time (s) ω (1/s)

FIG. 7. Stress relaxation modulus G(t) of SB40-7.13, fit by FIG. 9. Complex shear modulus G∗(ω) = G0(ω) + iG00(ω) of ∗ the stress relaxation modulus Glef(t) of an LEF. Modulus G SB40-7.13, fit by the complex shear modulus Glef(ω) of an and characteristic time τ are the fit’s two parameters. The LEF. The SB40-7.13 was in equilibrium with laboratory air. SB40-7.13 was in equilibrium with laboratory air.

TABLE II. Borosilicones were prepared from STPDMS flu- 40 ids of six different viscosities. The average molecular weights η(t) (MW) are approximations based on viscosity. When the 30 η (t) LEF trimethyl borate (TMB) concentration shown was added to 20 γ· = 0.000001 s -1 the STPDMS fluid and the mixture dried, the resulting sim- (MPa·s) ple borosilicone had the largest modulus G observed for that SB40-7.13 τ = 42 s η(t) 10 fluid. ηLEF = 30.0 MPa·s 0 Name Viscosity MW TMB G 30 (cSt) (Da) (wt%) (MPa) SB40-7.13 40a 850 7.13 1.366 ± 0.140 20 SB70-2.60 70b 2700 2.60 0.685 ± 0.036 b γ· = 0.0001 s -1 SB100-2.20 100 4000 2.20 0.562 ± 0.016

(MPa·s) b 10 τ = 36 s SB750-0.55 750 18000 0.55 0.221 ± 0.012

η(t) b η = 28.0 MPa·s SB2000-0.33 2000 36000 0.33 0.166 ± 0.008 LEF b 0 SB3500-0.28 3500 42500 0.28 0.146 ± 0.010 30 a Andisil OH fluid, AB Specialty Silicones b Masil SFR fluids, Dystar 20 γ· = 0.003 s -1 (MPa·s) 10 τ = 35 s

η(t) Other Simple Borosilicones ηLEF = 26.6 MPa·s 0 0 50 100 150 200 Simple borosilicones were also prepared from five Time (s) other STPDMS fluids of increasing viscosity and average molecular weight (Table II). For each viscosity, borosil- FIG. 8. Shear viscosity η(t) of SB40-7.13 at three different shear ratesγ ˙ . Each measurement is fit by the shear viscosity icones with slightly different TMB concentrations were ηlef(t) of an LEF. The SB40-7.13 was drier than laboratory prepared and studied to find the highest modulus. The air. stress relaxation moduli G(t) and shear viscosities η(t) of those peak-modulus simple borosilicones were then mea- sured and fit by Glef(t) and ηlef(t), respectively. largest modulus observed in the Andisil OH 40 borosili- The fits agree well with the data, lending support to cones. the idea that simple borosilicones are LEFs, and they yield values for the modulus G and characteristic time τ. Three measurements of SB40-7.13’s shear viscosity η(t) G decreases as the density of boron crosslinks decreases, are shown in Fig. 8, each fit by ηlef(t). Slight shear thin- consistent with rubber elasticity theory, and τ decreases ning is observed as the shear rate increases. The modulus with moisture content. G obtained by dividing ηlef by τ for many shear viscosity The measurements of complex shear modulus G∗(ω) measurements is G = 0.761 ± 0.219 MPa, somewhat less are far more interesting (Fig. 10). At small ω, each mea- than the modulus obtained by the sudden compression ∗ ∗ sured G (ω) agrees well with Glef(ω) from Eq. (28), con- technique. sistent with the simple borosilicone being a Lodge Elastic ∗ ∗ A measurements of SB40-7.13’s complex shear modu- Fluid. At large ω, however, G (ω) and Glef(ω) differ sig- ∗ ∗ nificantly. lus G (ω) is shown in Fig. 9, along with a fit by Glef(ω). The fit is excellent at small ω, but the observed loss mod- For simple borosilicones made from lower viscosity flu- ulus G00(ω) modestly exceeds the prediction at large ω. ids, SB40-7.13, SB70-2.60, and SB100-2.20, measurement 10

10 0 G'(ω) 10 -1 Spring GLEF '(ω) (au) -2 SB70-2.60 10 G = 1.09 au τ = 9.9 s G*(ω) 10 -3 Dashpot 10 -4 10 0 G''(ω) FIG. 11. The Maxwell model consists of a spring in series 10 -1 GLEF ''(ω) with a dashpot. (au) 10 -2 SB100-2.20 G = 1.04 au

G*(ω) 10 -3 τ = 11.6 s another by more than just boron crosslinks, a borosili- 10 -4 0 cone’s network is not purely transient and its stress-strain 10 behavior no longer obeys Eq. (4), the Lodge Equation. -1 10 Before examining more complicated borosilicones, a (au) 10 -2 SB750-0.55 more complete theory is needed. To apply to borosili- G = 0.46 au

G*(ω) -3 cones in which the strands are coupled to one another 10 τ = 5.7 s by more than just transient attachments, that extended -4 10 theory must be able to describe a fluid whose network 10 0 is intermediate between the transient network of an LEF -1 10 and the permanent network of an elastic solid. (au) 10 -2 SB2000-0.33 The most expeditious path to that more complete the- -3 G = 0.33 au G*(ω) ory involves viscoelastic modeling. Ordinary viscoelastic 10 τ = 5.9 s models can be found in any textbook on rheology or vis- 10 -4 10 0 coelastic materials[21, 39] and will be reviewed here only -1 briefly. They are generally built from combinations of 10 two simple conceptual elements: springs and dashpots. (au) -2 10 SB3500-0.28 A spring element exhibits an elastic relationship be- -3 G = 0.29 au G*(ω) 10 tween stress σ(t) and strain (t): τ = 4.4 s -4 10 σ(t) = G(t), (32) 10 -3 10 -2 10 -1 10 0 10 1 10 2 10 3 ω (1/s) where G is the spring element’s modulus. For steady iωt FIG. 10. Complex shear moduli G∗(ω) = G0(ω) + iG00(ω) of oscillatory strain (ω, t) = 0e , the stress is σ(ω, t) = iωt ∗ simple borosilicones SB70-2.60 through SB3500-0.28, each fit G0e and the spring element’s complex modulus Gs (ω) ∗ by the complex shear modulus Glef(ω) of an LEF. is

∗ σ(ω, t) Gs (ω) ≡ = G. (33) and theory disagree only in the loss modulus G00(ω), with (ω, t) the simple borosilicones exhibiting modest excess loss. A dashpot exhibits a viscous relationship between This excess is probably due to fast dissipative processes stress σ(t) and strain (t): not included in the TN model. For simple borosilicones made from higher viscosity flu- d σ(t) = η (t). (34) ids, SB750-0.55, SB2000-0.33, and SB3500-028, however, dt measurement and theory also disagree about the stor- where η is the dashpot’s viscosity. For steady oscillatory age modulus G00(ω), with the simple borosilicones ex- strain (ω, t) =  eiωt, the stress is σ(ω, t) = iωη eiωt hibiting modest excess storage. These excesses suggest 0 0 and the dashpot element’s complex modulus G∗(ω) is that fast dynamical processes, both dissipative and non- d dissipative, are present in long-chain borosilicones. σ(ω, t) G∗(ω) ≡ = iωη (35) d (ω, t)

VISCOELASTIC MODELS The Maxwell model While Green and Tobolsky’s transient network model and its Lodge Elastic Fluid predict the behaviors of sim- The Maxwell model for a viscoelastic fluid consists of ple borosilicones, they do less well with more complicated only two elements: a spring in series with a dashpot (Fig. ∗ borosilicones. Once its PDMS chains are coupled to one 11). Its complex modulus Gm(ω) can be calculated from 11

∗ ∗ Gs (ω) and Gd(ω) using a series formula analogous to that and that of a dashpot, Eq. (34), can be written as for electrical conductance: d1 −1   σ(t) = Gτ 1 (t), (41) ∗ 1 1 dt Gseries(ω) = ∗ + ∗ , (36) G1(ω) G2(ω) where τ ≡ η/G. Together, these relationships are ∗ ∗ where G1(ω) and G2(ω) characterize the individual ele- dn σ(t) = Gτ n (t), (42) ments. For the Maxwell model, dtn  −1 ∗ 1 1 where n = 1 for a dashpot and n = 0 for a spring. For Gm(ω) = ∗ + ∗ iωt Gs (ω) Gd(ω) steady oscillatory strain (ω, t) = 0e , the stress is iωt ∗ −1 σ(ω, t) = σ e and the complex modulus G (ω) is  1 1  0 n = + ∗ n G iωη Gn(ω) = G(iωτ) . (43) iωτ = G , (37) In ordinary calculus, the differentiation operator Dx is 1 + iωτ taken only to integer powers n where τ ≡ η/G. dn Since G∗ (ω) has the same form as G∗ (ω), the com- n m lef Dx ≡ n . (44) plex shear modulus of an LEF in Eq. (28), the Maxwell dx model perfectly characterizes the shear response of an In the branch of mathematical analysis known as frac- LEF. It accurately predicts the shear responses of the tional calculus, however, that differentiation operator can simple borosilicones, at least for small ω. be taken to any real or even complex power α The Maxwell model’s stress relaxation modulus Gm(t) dα can be obtained from G∗ (ω) via a simple transformation. Dα ≡ . (45) m x dxα For any fluid, G(t) can be obtained from G∗(ω) by divid- ing Eq. (27) by ω and then operating on both sides with Replacing the integer n in Eq. (42) with the real number R ∞ β results in a fractional-order stress-strain relationship the one-sided Fourier transform 0 dω sin(ωt): Z ∞ dβ 2 1 ∗ σ(t) = Gτ β (t) (46) G(t) = dω sin(ωt) Re [G (ω)] . (38) dtβ π 0 ω that defines a spring-pot. For β = 0, the spring-pot re- For the Maxwell model, duces to a spring. For β = 1, it reduces to a dashpot. For 2 Z ∞ 1 ω2τ 2 0 < β < 1, the spring-pot is a fractional-order viscoelas- Gm(t) = dω sin(ωt) G 2 2 π 0 ω 1 + ω τ tic element with behaviors intermediate between those of −t/τ a spring and those of a dashpot. = Ge . (39) α Any fractional differentiation operator Dx must repro- Since Gm(t) has the same form as Glef(t), the stress duce ordinary differentiation when α is integer. When relaxation modulus of an LEF in Eq. (6), the Maxwell α is non-integer, however, there is some flexibility in α model also characterizes the compression responses of an Dx . The operator used here is the Riemann-Liouville LEF and the simple borosilicones. Although the Maxwell left-sided derivative[10, 33], model is a one-dimensional concept, its constitutive equa- n Z x tion is the same as that of an LEF, Eq. (4). RL α 1 d n−α−1 Da+ [f(x)] ≡ n (x − ξ) f(ξ)dξ, Γ(n − α) dx a (47) The Fractional Maxwell model for the case in which n = 0 and a = −∞. Denoted α −∞Dx , this specific fractional derivative operator is The more complete theory, however, requires more 1 Z x Dα[f(x)] = (x − ξ)−α−1f(ξ)dξ. (48) than just springs and dashpots. It also requires a vis- −∞ x Γ(−α) coelastic element that is intermediate between a spring −∞ and a dashpot, a fractional-order viscoelastic element That operator can be used to find the spring-pot’s ∗ known as a spring-pot[19]. complex modulus Gβ(ω). For steady oscillatory strain iωt Springs and dashpots are viscoelastic elements of inte- (ω, t) = 0e , Eq. (46) is ger order, meaning that their stress-strain relationships β β  iωt involve integer-order time derivatives. A spring’s stress- σ(ω, t) = G0τ −∞Dt e strain relationship, Eq. (32), can be written as 1 Z t = G τ β (t − ξ)−β−1eiωξdξ 0 Γ(−β) d0 −∞ σ(t) = G (t) (40) β β iωt dt0 = G0τ (iω) e . (49) 12

where u(x) is the unit step function. Because of the step Spring function, the last term in Eq. (53) is zero and Z t −β 0 β−1 0 β 0 Gfm(t) = G − Gτ (t − t ) Eβ,β(−a(t − t ) )dt β Spring-pot 0 β
= GEβ −(t/τ) , (55) FIG. 12. The Fractional Maxwell model consists of a spring where Ref. [16], Eq. (5.1) has been used. in series with a spring-pot. Eα,γ (z) and Eα(z) are the Mittag-Leffler functions,

∞ The spring-pot’s complex modulus G∗ (ω) is X zk β E (z) = (56) α Γ(αk + 1) ∗ σ(ω, t) β k=0 Gβ(ω) ≡ = G(iωτ) . (50) ∞ (ω, t) X zk Eα,γ (z) = , (57) As it must, G∗ (ω) reduces to G∗(ω) when β = 1 and Γ(αk + γ) β d k=0 ∗ Gs (ω) when β = 0. When the Maxwell model’s dashpot is replaced with a generalized exponentials that reduce to ordinary expo- spring-pot (Fig. 12), the result is the Fractional Maxwell nentials when α = 1 and γ = 1. Correspondingly, the ∗ model[19, 37]. Consisting of a spring element in series Fractional Maxwell model’s complex modulus Gfm(ω) ∗ and stress relaxation modulus G (t) reduce to those of with a spring-pot element, its complex modulus Gfm(ω) fm can be calculated using Eq. (36), the ordinary Maxwell model when β = 1. !−1 The constitutive equation for a material that exempli- ∗ 1 1 fies the Fractional Maxwell model can be obtained from Gfm(ω) = ∗ + ∗ Gs (ω) Gβ(ω) Gfm(t) by substituting its derivative[16]  1 1 −1 ! = + ∂ G t − t0 β−1 t − t0 β G G(iωτ)β G (t − t0) = E − ∂t0 fm τ τ β,β τ (iωτ)β (58) = G β . (51) 1 + (iωτ) into Eq. (5), resulting in With its three parameters G, τ, and β, the FM model includes all possible choices of spring and spring-pot. Z t G t − t0 β−1 ↔σ(t) = The Fractional Maxwell model’s stress relaxation mod- −∞ τ τ ulus Gfm(t) can be obtained from its complex modulus β!  0  ↔−1 ∗ t − t 0 0 Gfm(ω) using Eq. (38), E − C (t, t )dt .(59) β,β τ 2 Z ∞ sin(ωt)  G(iωτ)β  Gfm(t) = dω Re β π 0 ω 1 + (iωτ) When β = 1, this constitutive equation reduces to Eq. G Z ∞ eiωt − e−iωt   1  (4), the Lodge Equation and the constitutive equation = dω 1 − β for the Maxwell model. π −∞ 2iω 1 + (iωτ) −β Z γ+i∞   Gτ st 1 1 = G − e ds −β β 2πi γ−i∞ s τ + s Random Assembly Model −β Z γ+i∞   Gτ −st 1 1 + e ds −β β . (52) 2πi γ−i∞ s τ + s The Fractional Maxwell (FM) model has been intro- where s ≡ iω and γ is a positive real infinitesimal. The duced because, when the PDMS chains of a borosili- two integrals are inverse Laplace transforms, cone are coupled by more than just boron crosslinks, the   borosilicone’s network is no longer purely transient and −β −1 1/s the Lodge equation and Maxwell model no longer apply. Gfm(t) = G − Gτ L ; t a + sβ Such borosilicones will be called “non-simple” because  1/s  they have characteristics intermediate between those of +Gτ −βL−1 ; −t , (53) a + sβ elastic fluids with their temporary networks and those of elastic solids with their permanent networks. where a = τ −β. As shown in Ref. [26], Eq. (2.2.21), This intermediate territory is the domain of fractional-   Z t −1 1/s 0 0 β−1 0 β 0 order viscoelastic models and fractional-order elements L β ; t = u(t )(t−t ) Eβ,β(−a(t−t ) )dt , a + s 0 such as the spring-pot. Since the Maxwell model success- (54) fully predicts the behaviors of simple borosilicones, it is 13

(a) (b) (a)shear (b) shear shear z^

z^ y^ x^ y^ x^ (c) FIG. 14. (a) When cuboids extend alongz ˆ, they act in parallel shear under shear. (c) When cuboids extend alongx ˆ, they act in series.

face to surface (Fig. 14b), then those cuboids would act in series and the slice’s complex shear modulus would be ∗ −1 −1 −1 FIG. 13. (a) A random assembly of upright rectangular Gseries(ω) = β(Giωτ) + (1 − β)(G) . (61) cuboids (edges alongx ˆ,y ˆ, andz ˆ). (b) When surfaces per- pendicular toz ˆ are sheared alongx ˆ, there are no interactions Writing the parallel and series formulas in this fashion in they ˆ direction and the random assembly can be reduced allows them to be expressed together as to (c) a thin slice parallel to the xz plane. ∗ ν ν ν Gν (ω) = β(Giωτ) + (1 − β)(G) , (62) where ν = 1 for parallel and ν = −1 for series. reasonable to suppose that the FM model, with a spring- Of course, the slice’s actual cuboid structure is far more pot replacing the Maxwell model’s dashpot, will similarly complicated, with cuboids that are beside one another in predict the behaviors of non-simple borosilicones. the ˆxdirection acting in parallel and cuboids that are To begin building a case for the FM model, consider above one another in the ˆzdirection acting in series. Be- the microscopic structure of a non-simple borosilicone. cause the cuboids are arranged randomly, parallel and It has both transient and non-transient couplings be- series connections are equally abundant. tween its chains and those interactions produce viscous Following the mixing method of Lichtenecker[23], and elastic behaviors, respectively[31]. The material can ∗ Gra(ω) can be obtained by evaluating Eq. (62) in the be modeled as a randomly-distributed mixture of viscous limit ν → 0, where series and parallel connections are regions with complex shear modulus Giωτ and elastic equally common. Expanding zν near ν = 0 as regions with complex shear modulus G. Regardless of a region’s shape, it can be deconstructed zν = 1 + ln(z)ν + O(ν2), (63) into a collection of upright rectangular cuboids (rectan- the limit of Eq. (62) as ν → 0 is gular cuboids with edges along ˆx,ˆy,and ˆz). The entire ∗ material can thus be modeled as a random assembly of ln(Gra(ω)) = βln(Giωτ) + (1 − β)ln(G). (64) viscous and elastic cuboids (Fig. 13a). A parameter β Exponentiating both sides of Eq. (64) gives establishes the volume fractions of viscous cuboids (β) ∗ β 1−β and elastic cuboids (1 − β). Gra(ω) = (Giωτ) · G Although the random assembly is three-dimensional, = G(iωτ)β. (65) when a surface perpendicular to ˆzis sheared in the ˆxdi- ∗ rection (Fig. 13b), its complex shear modulus G∗ (ω) Gra(ω) has the same form as the complex modulus ra ∗ depends only on its structure in the xz plane. This ob- of a spring-pot Gβ(ω), given in Eq. (50), so a random servation assumes affine deformation and large-scale ho- assembly of elastic and viscous cuboids acts as a spring- mogeneity of the model, so that there are no interactions pot. Moreover, the parameter β is both the fractional in the ˆydirection and a thin slice cut parallel to the xz concentration of viscous cuboids in the random assem- plane (Fig. 13c) has the same complex shear modulus bly and the fractional-order of the spring-pot’s stress- as the entire model. If the slice is cut thin enough, it strain relationship, Eq. (46). These results support the will be uniform in they ˆ direction and thus effectively idea that a non-simple borosilicone’s mixture of transient two-dimensional. and non-transient chain couplings can produce fractional- If all the slice’s cuboids extended in the ˆzdirection order viscoelastic behavior. from surface to surface (Fig. 14a), then those cuboids would act in parallel and the slice’s complex shear mod- Random Network Model ulus would be

∗ 1 1 1 Most viscoelastic models are one-dimensional, so it Gparallel(ω) = β(Giωτ) + (1 − β)(G) . (60) would be nice to find a one-dimensional analog to the ran- If all of its cuboids extended in the ˆxdirection from sur- dom assembly model. Fortunately, the random assembly 14

(a) (b) an evenly-spaced square array of elements with appro- priate couplings. Its structure would then look similar C a G * a G * A A A C C to the random networks of capacitors and resistors used to model complex conductivity and dielectric response in a G * a G* materials[2, 8, 23, 44, 48]. D B B D D B Since the random assembly behaves as a spring-pot, so does the corresponding random network. Despite being composed entirely of dashpots and springs, this random network acts as a fractional-order viscoelastic element. While fractional-order viscoelastic elements a G* a G * Q S Q Q S S have been constructed previously out of integer-order elements[18, 36, 37], those constructions have involved * * highly ordered arrangements such as infinite ladders, aRGR aTGT R T trees, and fractals. For thermodynamic reasons, such orderly structures are unlikely to appear in real materi- als. Moreover, their fractional order β is determined by FIG. 15. (a) Two 4-cuboid portions of random assembly are network structure rather than a continuous parameter, transformed into (b) two 4-element portions of random net- usually limiting β to a small number of discrete values. ∗ work. The modulus of an element is the modulus Gcuboid In contrast, the random assembly and random network of its cuboid’s material times the aspect ratio acuboid of its models discussed here use randomness and statistics to cuboid. Series pairs AB and CD combine in parallel. Parallel pairs QS and RT combine in series. their advantage, so they should be easier to realize in real materials. Moreover, the fractional order β that charac- terizes these models can easily take any value 1 ≥ β ≥ 0. model transforms easily into a one-dimensional random network model. This transformation begins with a thin The Computational Random Network slice of the random assembly, uniform in they ˆ direction, and replaces each three-dimensional viscoelastic cuboid The Lichtenecker mixing method used to obtain the with a one-dimensional viscoelastic element. The com- random assembly’s complex shear modulus analytically is plex modulus of that element is the cuboid’s complex simple and elegant, but also rather mysterious. To verify shear modulus multiplied by the cuboid’s aspect ratio its results and obtain further insight into these models, a = width/height. A viscous cuboid is thus replaced by the random network model was studied computationally. a dashpot with complex modulus aGiωτ and an elastic The computational approach begins by constructing cuboid is replaced by a spring with complex modulus aG. a thin random assembly, using a recursive procedure to The viscoelastic elements are then coupled together in cut and recut a square region of the xz plane into bil- series or parallel, using the slice of random assembly as lions of tiny rectangles. The inconsequentialy ˆ dimension a guide. Elements replacing cuboids that are above one is omitted for simplicity. The rectangles are then ran- another in thez ˆ direction couple in series, while elements domly assigned viscous or elastic complex shear moduli, replacing cuboids that are beside one another in thex ˆ di- weighted statistically to ensure that the volume fraction rection couple in parallel. The two 4-cuboid portions of of viscous rectangles is β and of elastic rectangles is 1−β. the random assembly shown in Fig. 15a are thus trans- Each rectangle is then replaced by a viscoelastic ele- formed into the two 4-element portions of the random ment with a complex modulus that is the rectangle’s com- network shown in Fig. 15b. Cuboids/elements A and B plex shear modulus multiplied by its aspect ratio. For couple in series, C and D couple in series, and the coupled computational purposes, that element’s complex modu- pairs then couple in parallel. Similarly, Q and S couple lus is represented by its values at 200 discrete angular in parallel, R and T couple in parallel, and the coupled frequencies covering 20 orders of magnitude in ω. For pairs then couple in series. simplicity, τ ≡ 1 s. The overall random network is a vast hierarchical lat- As the recursive procedure returns up through its many ticework of coupled subnetworks that are themselves cou- levels, its viscoelastic elements are coupled and recou- pled subnetworks. Despite its complexity, this random pled to form a hierarchy of subnetworks that eventually network can be drawn in two dimensions without its become the entire random network. Any pair of sub- coupling lines crossing. The density of its elements is networks formed by cutting a rectangle along thez ˆ di- non-uniform, a consequence of the non-uniform number- rection is coupled in parallel. Any pair formed by cut- density of the cuboids they replace. Since one element ting a rectangle along thex ˆ direction is coupled in series. can always be divided into two half-elements in series or Each coupling is performed computationally on the two in parallel, however, the model could be expanded into complex-value arrays representing its subnetworks, pro- 15

(a) 10 2 1 (a) β = 0.100 ± 0.001 10 τ = 1.078 ± 0.074 s (au) 10 0 -1 G*(ω) 10

10 -2 10 6 3 (b) β = 0.505 ± 0.003 10 τ = 0.963 ± 0.081 s (au) 10 0 (b) G' cRN (ω) -3 G*(ω) 10 G' β(ω)

10 -6 10 10 5 (c) β = 0.908 ± 0.004 10 τ = 1.137 ± 0.076 s (au) 10 0 G''cRN (ω) -5 G*(ω) 10 G''β(ω) 10 -10 10 -10 10 -5 10 0 10 5 10 10 ω (1/s) FIG. 16. (a) A fraction of the rectangles in a random assembly with β = 0.5. Black represents viscous material and white FIG. 17. The computational random network model’s com- represents elastic material. (b) A fraction of the dashpots ∗ plex modulus Gcrn(ω) for (a) β = 0.1, (b) β = 0.5, and (c) and springs in a random network with β = 0.5. β = 0.9, with τ = 1 s and G = 1. Also shown are fits to the ∗ spring-pot’s complex modulus Gβ (ω), along with values of β and τ obtained from those fits. ducing a single array that is returned to the next higher level. The final coupling yields a complex-value array representing the complex modulus of the entire random directions (e.g., alongx ˆ at even recursion levels, alongy ˆ network. at odd levels) resulted in complex moduli that did not A recursion depth of 36 levels gives clear, compelling resemble that of a spring-pot. In contrast, the choice results. Since each level of recursion divides a rectangle of cut amount did not have to be random. Whether a into two smaller rectangles, 36 levels of recursion pro- rectangle was cut into equal or random subrectangles had duces a random assembly with 236 = 68, 719, 476, 736 no significant effect on the overall complex modulus. The rectangles. The corresponding random network has data in Fig. 17 were produced by cutting rectangles into 68,719,476,736 viscoelastic elements. Figure 16 displays equal subrectangles. (a) a small fraction (1.3 · 10−7) of the random assembly with β = 0.5 and (b) an even smaller fraction (8 · 10−10) of the random network. Spring-pot to Fractional Maxwell model Figure 17 shows the computational random network’s ∗ Although the spring-pot arises naturally when both complex modulus Gcrn(ω) for (a) β = 0.1, (b) β = 0.5, and (c) β = 0.9, with its real and imaginary parts sepa- transient and non-transient couplings are included in the 0 RA and RN models, the spring-pot’s complex modulus rated into its storage modulus Gcrn(ω) and its loss mod- 00 diverges when ωτ → ∞ for β > 0. The spring-pot alone is ulus Gcrn(ω), respectively. Also shown are fits by the ∗ therefore not a realistic viscoelastic model for non-simple spring-pot’s complex modulus Gβ(ω), again separated into storage and loss moduli, borosilicones. One way to avoid the divergence is to reconsider the 0 β Gβ(ω) = G(ωτ) cos(πβ/2) (66) materials used in the random assembly. Since a simple 00 β borosilicone exhibits Maxwell model behavior, perhaps Gβ(ω) = G(ωτ) sin(πβ/2). (67) the random assembly should contain a Maxwell material The fits are nearly perfect over 15 orders of magnitude ∗ Gm(ω) rather than a viscous material. Making that sub- in ω and the values of β and τ obtained from those fits stitution and applying the Lichtenecker mixing method are almost identical to those used when constructing the to the modified random assembly yields complex shear computational random network. modulus It was observed that when constructing the random assembly, the choice of cut direction, whether alongx ˆ or  iωτ β G∗ (ω) = G . (68) zˆ, had to be random. Any non-random sequence of cut mra 1 + iωτ 16

∗ Gmra(ω) does indeed converge when ωτ → ∞. More- temperature, and releases an acetic acid molecule. Its ∗ ∗ over, it reduces to Gm(ω) when β = 1 and Gs (ω) when second acetoxy group forms a crosslink more slowly. Its β = 0. But despite these promising characteristics, third and final acetoxy group can take hours or days to ∗ Gmra(ω) fails the most important test: it does not fit the form a crosslink, even at elevated temperature, and may measured complex shear moduli of non-simple borosili- not form a crosslink at all if the mixture is not heated cones. long enough. Another way to avoid the divergence is to recognize From work on simple borosilicones, it is estimated that the finite compressibility of real liquids. A physically 5.8 wt% VTAS is stoichiometric saturation in 70 cSt reasonable model for an ordinary liquid requires a com- STPDMS. Since the gelation threshold for a trifunctional pressible spring in series with a dashpot, resulting in the crosslinker is 50% of saturation, 3 wt% VTAS should gel Maxwell model. Correspondingly, a physically reason- 70 cSt STPDMS. In practice, however, VTAS forms its able model for a non-simple borosilicone requires a com- third crosslink so slowly that higher VTAS fractions are pressible spring in series with a spring-pot, resulting in needed to approach the gelation threshold in a reasonable the Fractional Maxwell model. amount of time. ∗ As desired, the FM model’s complex modulus Gfm(ω), The non-simple borosilicone was therefore made by given in Eq. (51), converges when ωτ → ∞. It also adding 4.5 wt% VTAS to SFR 70. After blending, the ∗ ◦ reduces to Gm(ω) when β = 1 and, after first taking mixture was heated to 130 C for 5 hours to promote ∗ the limit as ωτ → ∞, to Gs (ω) when β = 0. More crosslinking and evaporation of acetic acid. The result- ∗ importantly, however, Gfm(ω) passes the critical test: it ing partially crosslinked silicone fluid (PCS1) had a room- fits the measured complex shear moduli of non-simple temperature viscosity of 1.12 Pa·s, 16 times that of the borosilicones. original fluid (SFR 70). PCS1 was vacuum dried to remove residual acetic acid, converted to a borosilicone by adding 1.5 wt% TMB, and NON-SIMPLE BOROSILICONES vacuum dried again. It was pressed into a sheet and allowed to equilibrate with laboratory air for 10 days. Non-simple borosilicones have both transient and non- The resulting non-simple borosilicone, designated NB70- transient couplings between their PDMS chains. The PCS1, is a stiff, transparent liquid with an extremely transient couplings are temporary (boron) crosslinks, high viscosity. but the non-transient couplings can vary. The clas- sic non-transient coupling is a permanent crosslink, so the quintessential non-simple borosilicone is a PDMS Stress Relaxation Modulus network with a mixture of temporary and permanent crosslinks. When those crosslinks are randomly dis- The stress relaxation modulus G(t) of NB70-PCS1 was tributed, the network is a physical realization of the RA measured using the sudden compression technique. One and RN models. such measurement is shown in Fig. 18a, along with a One way to form those permanent crosslinks is through fit by Gfm(t), the Fractional Maxwell model’s stress re- a condensation reaction with a silane crosslinker. Un- laxation modulus given in Eq. (55). The quality of the fortunately, most silane crosslinkers require catalysts fit indicates that the Fractional Maxwell model does in- that remain in the borosilicone and affect its properties. deed describe the stress-strain behavior of NB70-PCS1, Acetoxysilanes (=Si-CH3COOH), however, will crosslink at least for sudden compression. STPDMS fluids at elevated temperatures without cata- The fit provides values for the model’s three param- lysts. Non-simple borosilicones were therefore produced eters: the modulus G, the fractional order β, and the using vinyltriacetoxysilane (VTAS) as the permanent characteristic time τ. The modulus G was found to be crosslinker. 0.852±0.038 MPa, somewhat larger than the modulus of To avoid the extreme stiffness and brittleness of SB70-2.60. That result indicates that the presence of per- borosilicones based on 40 cSt STPDMS (Andisil OH manent crosslinks in the material stiffens its time-average 40), non-simple borosilicones were instead based on 70 network as compared to a network having purely tempo- cSt STPDMS (Dystar Masil SFR 70). The permanent rary crosslinks. The characteristic time τ was found to crosslinks were formed by adding VTAS to SFR 70 and be 3.6 s, but increased slowly over time as acetic acid heating the mixture to at least 100 ◦C. Although the ex- from the VTAS gradually dissipated. tent of permanent crosslinking was determined primarily β is the fractional order of the FM model’s spring-pot by the VTAS concentration, the temperature and dura- and also the random assembly’s volume concentration of tion of the heating were also quite important. viscous cuboids. The fit gives β = 0.849 ± 0.014, cor- While VTAS is a trifunctional crosslinker, its reactivity responding to a random assembly that is 85% viscous decreases as its coordination increases. Its first acetoxy and 15% elastic. The effect of β < 1 is to prolong the group reacts quickly with a silanol group, even at room stress relaxation so that it is slower-than-exponential, as 17

10 0 12 η(t) 10 (a) FM Model G(t) η (t) -1 NB70-PCS1 G (t) 8 FM 10 FM γ· = 0.00001 s -1

(MPa) G = 0.88 MPa 6 G = 0.33 MPa -2 β = 0.840 (MPa·s) NB70-PCS1 G(t) 10 4 β = 0.846

τ = 3.6 s η(t) 2 τ = 14 s -3 10 0 10 0 8 (b) Maxwell Model 6 10 -1 G = 0.88 MPa γ· = 0.0001 s -1 (MPa) G (t), τ = 50.6 s 4 -2 M G = 0.38 MPa (MPa·s) G(t) 10 G (t), τ = 3.6 s M β = 0.871

η(t) 2 τ = 8 s 10 -3 0 0 10 8 G (t) (c) Stretched Exponential SE 10 -1 6 · -1 G = 1.90 MPa γ = 0.001 s (MPa) -2 β = 0.261 4 G = 0.40 MPa

10 (MPa·s) G(t) τ = 2.3 s β = 0.828

η(t) 2 τ = 6 s 10 -3 10 -2 10 -1 10 0 10 1 10 2 10 3 0 0 500 1000 1500 Time (s) Time (s) FIG. 18. (a) Stress relaxation modulus G(t) of NB70-PCS1, FIG. 19. Shear viscosity η(t) of NB70-PCS1 at three different fit by the FM model’s stress relaxation modulus G (t). Mod- fm shear ratesγ ˙ . Each measurement is fit by the FM model’s ulus G, fractional order β, and characteristic time τ are the shear viscosity η (t). Modulus G, fractional order β, and fit’s three parameters. Fits by (b) the Maxwell model’s stress fm characteristic time τ are the fit’s three parameters. relaxation modulus Gm(t) and (c) a stretched exponential modulus Gse(t) do not match the data well.

route is to substitute Gfm(t) into Eq. (15), Z t  0 β! observed in the measurement. For comparison, the expo- t 0 ηfm(t) = GEβ − dt nential relaxation predicted by the Maxwell model does 0 τ not fit the data (Fig. 18b).  β! Since the 1854 work of Kohlrausch[20], relaxation pro- t = GtEβ,2 − , (69) cesses that are slower than exponential have often been fit τ −(t/τ)β by stretched exponentials of the form Gse(t) = Ge . where Ref. [26], Eq. (2.3.17) has been used to do the Such fits are mostly empirical because few theories pre- integral. dict stretched exponential relaxation[32]. A connection When β = 1, ηfm(t) reduces to ηlef(t) given in Eq. between theories and the non-simple borosilicones could (16). With its only time-dependent term a decaying ex- not be found, which is just as well because Gse(t) does ponential, η (t) approaches η = Gτ at long times. not fit the NB70-PCS1 data (Fig. 18c). lef lef For 1 > β > 0, however, ηfm(t) does not approach a constant value at long times. Instead, it behaves as τ β Shear Viscosity η (t) = Gt1−β + O(t1−2β), (70) fm Γ(2 − β) The time-dependent shear viscosity η(t) of NB70-PCS1 where Ref. [33], Eq. (1.143) has been used. The time- was measured over a range of shear rates using the linear dependent viscosity ηfm(t) of a Fractional Maxwell ma- shear technique. Because the results differ qualitatively terial thus increases as Ct1−β forever. from those observed in simple borosilicones, they cannot Measurements of NB70-PCS1’s shear viscosity at three be explained by the transient network and Maxwell mod- different constant shear ratesγ ˙ are shown in Fig. 19, els. Instead, they required the FM model and an analy- along with fits by ηfm(t). Those fits yield values for sis of how materials exemplifying that model respond to the modulus G, the fractional order β, and the char- constant shear strain. acteristic time τ. Together, the measurements give β = As before, the analysis assumes that surfaces perpen- 0.848 ± 0.035, similar to the β obtained by the sudden dicular to axis 2 are sheared along axis 1, so shear stress compression technique. As predicted by the FM model, σs(t) = σ21(t). The calculation could proceed using the the measured shear viscosities η(t) increased with t for appropriate constitutive equation, Eq. (59), but a faster as long as the measurements could proceed. 18

0 0 10 10 G'(ω) G''(ω) (a) Maxwell Model GM'(ω)

(au) G' (ω) G'' (ω) -1 NB70-PCS1 FM FM 10 G ''(ω) 10 -1 M G = 1.63 au -2 2 G*(ω) β = 0.804 = -2 10 slope = 1 10 τ = 3.6 s slope 10 -3 -3 -2 -1 0 1 2 3 10 10 10 10 10 10 10 10 0 ω (1/s) (b) FM Model GFM '(ω) 10 -1 G ''(ω) FIG. 20. The complex shear modulus G∗(ω) = G0(ω)+iG00(ω) β transition FM G*(ω)-2 slope G*(ω) = β = 0.8 of NB70-PCS1, fit by the FM model’s complex shear modulus 10 β G∗ (ω). fm slope = 10 -3 10 -3 10 -2 10 -1 10 0 10 1 10 2 10 3 Complex Shear Modulus ω (1/s)

∗ FIG. 21. (a) The Maxwell model’s complex modulus Gm(ω) 0 2 00 A measurement of the complex shear modulus for for G = 1 and τ = 1 s. As ω → 0, Gm(ω) ∝ ω and Gm(ω) ∝ ∗ NB70-PCS1 is shown in Figure 20, along with a fit by ω. (b) The FM model’s complex modulus Gfm(ω) for G = 1, ∗ β = 0.8, and τ = 1 s. As ω → 0, G0 (ω) ∝ ωβ and G00 (ω) ∝ Gfm(ω). The fit is excellent and provides values for the fm fm β 1/β 0 fractional order β = 0.804 and the characteristic time ω . In the transition region cos (πβ/2) < ωτ < 1, Gfm(ω) β τ = 3.6 s. This β is slightly smaller than that obtained changes more rapidly than ω . via sudden compression or constant shear. Despite their minor differences, the three measurement 0 10 G''(ω) G'(ω) techniques agree that NB70-PCS1 is about 15 to 20% of GFM ''(ω) GFM '(ω)

(au) NB70-TFS the way along the path from viscous behavior to elastic -1 10 G = 1.65 au behavior. It is definitely a fractional-order viscoelastic β = 0.892 material. G*(ω) = 13.6 s 10 -2 τ

10 -3 10 -2 10 -1 10 0 10 1 10 2 10 3 ω (1/s) Other Non-Simple Borosilicones FIG. 22. The complex shear modulus G∗(ω) = G0(ω)+iG00(ω) Non-simple borosilicones can be differentiated from of NB70-TFS, fit by the FM model’s complex shear modulus 0 00 ∗ simple borosilicones by the behaviors of G (ω) and G (ω) Gfm(ω). at small ωτ. In a simple borosilicone described by the Maxwell model, G0(ω) ∝ ω2 and G00(ω) ∝ ω as ω → 0. In a log-log plot, the slope of G0(ω) is 2 and the slope of ω, apart from some excess loss at high frequencies. G00(ω) is 1 (Fig. 21). Fumed silica (FS) and its chemically treated version In a non-simple borosilicone described by the FM (TFS) are routinely dispersed in silicone rubbers to en- model, G0(ω) ∝ ωβ and G00(ω) ∝ ωβ as ω → 0. There is hance tensile strength, tear strength, and toughness. The also a transition region cos1/β(πβ/2) < ωτ < 1 in which reinforcing nature of these nanometer-scale amorphous G0(ω) changes more rapidly than ωβ. In a log-log plot, silica particles stems from their physical and chemical the slopes of G0(ω) and G00(ω) are β as ω → 0, but the interactions with the silicone chains. In borosilicones, slope of G0(ω) exceeds β in the transition region. however, fumed silica evidently plays another important ∗ ∗ Non-simple borosilicones based on partially- role. The fit of G (ω) by Gfm(ω) suggests that some frac- crosslinked silicones (e.g., NB70-PCS1) clearly exhibit tion of the particle-chain interactions act as non-transient these reduced asymptotic slopes, but they are not alone. couplings so that the RA model applies and NB70-TFS Borosilicones containing reinforcing fillers or silicone exhibits FM model behavior with β ≈ 0.9. gums also exhibit reduced asymptotic slopes and are Fig. 23 shows G∗(ω) for NB70-GUM, a borosilicone thus non-simple. made by adding 50 wt% 20,000,000 cSt PDMS gum Fig. 22 shows G∗(ω) for NB70-TFS, a borosilicone (Gelest DMS-T72) to SB70-2.60. The gum was added made by adding 40 wt% Treated Fumed Silica (Cabot to the SFR 70 fluid before adding the TMB and vacuum TS-530) to SB70-2.60. The silica was milled into the drying. The slopes of G0(ω) and G00(ω) are significantly SFR 70 fluid before adding the TMB and vacuum drying. reduced as ω → 0, consistent with FM model behavior Consistent with the FM model, G0(ω) and G00(ω) exhibit with β ≈ 0.9. For small ω, G∗(ω) is fit fairly well by ∗ reduced slopes as ω → 0. Moreover, the FM model pro- Gfm(ω). For large ω, however, the fit is poor. vides an excellent fit over the entire measured range of Like fumed silica, silicone gum strengthens silicone 19

10 0 8 G'(ω) 10 -1 7 (au) NB70-GUM GFM '(ω) 10 -2 G = 0.62 au = 0.893 G''(ω) 6 X

G*(ω) β -3 10 τ = 1.0 s GFM ''(ω) 5 -2 -1 0 1 2 3 10 10 10 10 10 10 X ω (1/s) 4 X . ηE(t) ϵ ηE,FM (t) (MPa·s)

∗ 0 00 E -1 FIG. 23. The complex shear modulus G (ω) = G (ω)+iG (ω) η 3 0.40 s of NB70-GUM, fit by the FM model’s complex shear modulus 0.30 s-1 ∗ -1 Gfm(ω). 2 0.25 s 0.22 s-1 1 0.20 s-1 rubbers and borosilicones. A gum’s long polymer chains 0 disperse stress through their entanglements and extended 0 1 2 3 4 5 6 7 8 9 reaches. When woven into a borosilicone, the gum’s Time (s) chains evidently act as non-transient couplings, so that the RA model applies and results in approximately FM FIG. 24. Time-dependent elongation viscosity ηe(t) curves model behavior at small ω. At large ω, however, excess for NB70-TFS. ηe(t) was measured at the constant elonga- loss first seen in the high-MW borosilicones (Fig. 10) tion rates ˙ indicated. Breaks of the sample are marked by obscure much of the FM model behavior. x. ηe,fm(t) was calculated numerically from Eq. (71), using G = 1.05 MPa, β = 0.893, and τ = 0.796 s, values obtained by fitting a separate compression measurement of the same NB70-TFS. Elongation Viscosity

With its fumed silica reinforcement, NB70-TFS toler- where Ref. [16], Eq. (7.1) has been used. The first ates considerable tensile stress before breaking and its integral diverges with a leading term approximated using time-dependent elongation viscosity ηe(t) can be mea- Ref. [33], Eq. (1.143) as sured at interesting elongation rates. Figure 24a shows 2Gτ β its ηe(t) measured at five constant-elongation rates ˙ from 1−β t 1F1(1 − β; 2 − β; 2˙t), (73) 0.20 s−1 to 0.40 s−1, using the exponential stretching Γ(1 − β)(1 − β) technique described earlier. The tensile strength of NB70-TFS is approximately where 1F1(a; b; z) is the Kummer confluent hypergeomet- 1.5 MPa and the beam broke suddenly whenever its ten- ric function. The elongation viscosity ηe,fm(t) therefore sile stress exceeded that value. Breakage occurred dur- grows without limit as t → ∞ and any borosilicone de- ing elongation at the three highest elongation rates. In scribed by the FM model must eventually break when each break, the beam snapped near its midpoint and left subject to constant-rate elongation. smooth fracture faces, a behavior familiar to anyone who For finite t, ηe,fm(t) could only be computed numeri- has stretched silicone bouncing putty (e.g., Silly Putty) cally. Unfortunately, fits using the computational ηe,fm(t) quickly enough to break it. were unsuccessful, so the model’s parameters G = 1.05 The FM model’s time-dependent elongation viscosity MPa, β = 0.893, and τ = 0.80 s were found instead using the sudden compression technique on the same portion ηe,fm(t) was calculated for comparison. For constant-rate of NB70-TFS. The ηe,fm(t) curves shown in Fig 24 were elongation ˙, starting at t = 0, ηe,fm(t) can be obtained then calculated numerically at the five constant elonga- by substituting Gfm(t) into Eq. (21), tion rates ˙. Z t 0 β 2t ˙ 0 0 The agreement between ηe(t) and ηe,fm(t) is far from ηe,fm(t) = 2G Eβ(−(t /τ) )e dt perfect, with initial slopes differing by about a factor of 2. 0 Z t Nonetheless, measurement and model exhibit significant 0 β −t˙ 0 0 +G Eβ(−(t /τ) )e dt . (71) similarities in scale and curvature, particularly given that 0 the ηe,fm(t) curves involved no parameter adjustments. Closed form solutions for these integrals could not be found, but it was possible to determine their behaviors as t → ∞. The second integral converges to a finite value, Revisiting Simple Borosilicones

(˙τ)β−1 While simple borosilicones exemplify the Lodge Elastic Gτ , (72) 1 + (˙τ)β Fluid concept and their measured complex shear moduli 20

0 FM model should reduce to the Maxwell model. That 10 G'(ω) -1 G '(ω) outcome is not observed. Instead, the fits shown in Fig. 10 SB40-7.13 FM

(au) 25 consistently yield β just less than 1 and, in four of 10 -2 G = 1.99 au β = 0.976 ± 0.007 the six fits, the 95% confidence interval excludes β = 1. -3 G*(ω) 10 τ = 7.3 s There appear to be small but non-zero concentrations 10 -4 of non-transient couplings in simple borosilicones, due 10 0 G''(ω) perhaps to steric hindrances, impurities, or clustering of 10 -1 G ''(ω) boron and oxygen atoms. SB70-2.60 FM (au) 10 -2 G = 1.10 au β = 0.992 ± 0.007 -3 G*(ω) 10 τ = 9.2 s The Approach to Solid 10 -4 0 10 The RA and FM models both predict liquid behavior, ∗ 10 -1 G (0) = 0 and G(∞) = 0, when β > 0. One might there- SB100-2.20 (au) 10 -2 G = 1.07 au fore expect that, as the fraction of permanent crosslinks β = 0.971 ± 0.006 -3 in a non-simple borosilicone increases almost to one, the G*(ω) 10 τ = 10.1 s borosilicone will remain liquid and its β will decrease al- 10 -4 most to zero. 0 10 In reality, however, when β drops below about 0.8, 10 -1 something not predicted by the RA or FM models occurs:

(au) SB750-0.55 ∗ -2 the borosilicone begins to exhibit solid behavior, G (0) 6= 10 G = 0.46 au -3 β = 0.985 ± 0.019 0 and G(∞) 6= 0. The RA and FM models, which do G*(ω) 10 τ = 5.2 s not consider real crosslinks between real polymer chains, 10 -4 fail to predict the polymer phase transition known as 0 10 gelation. 10 -1 A simple borosilicone is a network liquid composed of (au) -2 SB2000-0.33 silicone chains connected only by temporary crosslinks. 10 G = 0.33 au When a few permanent crosslinks are added, they cre- G*(ω) 10 -3 β = 0.984 ± 0.025 τ = 5.5 s ate some permanently-crosslinked subnetworks—that is, -4 10 parts of the overall network that cannot come apart. 0 10 As long as the permanent crosslink concentration is 10 -1 low, those subnetworks are small and isolated, and con-

(au) SB3500-0.28 -2 nect to one another only through temporary crosslinks. 10 G = 0.29 au -3 β = 0.967 ± 0.031 The borosilicone is a sol (or solution) of permanently- G*(ω) 10 τ = 3.8 s crosslinked subnetworks in an otherwise temporarily- -4 10 crosslinked liquid borosilicone. 10 -3 10 -2 10 -1 10 0 10 1 10 2 10 3 ω (1/s) Once the concentration of permanent crosslinks ex- ceeds the gelation threshold, however, the permanently- FIG. 25. Complex shear moduli G∗(ω) = G0(ω) + iG00(ω) of crosslinked subnetworks are no longer isolated. In- simple borosilicones SB70-2.60 through SB3500-0.28, each fit stead, many or most of them coalesce into an enormous ∗ by the FM model’s complex shear modulus Gfm(ω). The fit’s permanently-crosslinked network that spans the material 95% confidence interval is given for β. and prevents it from fully relaxing stress, even as t → ∞. The borosilicone is then a gel and a network solid. Its permanent crosslinks give a fixed aspect to its overall net- are well fit by the Maxwell model’s complex shear mod- work topology and its temporary crosslinks add a tem- ∗ ulus Gm(ω) (Fig. 10), the same data can be fit even porary aspect as well. In effect, this type of borosilicone more tightly using the Fractional Maxwell model’s com- is a network liquid piggybacking on the framework of a ∗ plex shear modulus Gfm(ω) (Fig. 25). Of course, adding network solid. another parameter to a fit can improve that fit even when To find the fractional order β associated with the gela- the added parameter has no physical justification. In tion threshold, a liquid borosilicone was produced as close this case, however, including the fractional order β is a as possible to that threshold. Designated NB70-PCS2, it recognition that simple borosilicones are not perfect sys- was based on PCS2, a nearly-gelled partially-crosslinked tems and could have some non-transient couplings be- silicone fluid. tween their polymer chains. PCS2 was made by adding 4.0 wt% VTAS to SFR If there were no non-transient couplings in the simple 70 and heating the mixture to 150 ◦C until it had al- borosilicones, the fits to data should find β = 1 and the most gelled. When, after more than 30 hours, the liquid 21

10 10 0 G(t) G'(ω) G''(ω) NB70-PCS2

-1 G (t) (au) G' (ω) G'' (ω) 10 FM FM FM G = 0.65 MPa 10 -1 NB70-PCS2 β = 0.774 G = 1.33 au (MPa)

-2 τ = 61 s G*(ω) -2 β = 0.790 G(t) 10 10 τ = 28 s

-3 -2 -1 0 1 2 3 10 -2 10 -1 10 0 10 1 10 2 10 3 10 4 10 10 10 10 10 10 10 (1/s) Time (s) ω

∗ 0 00 FIG. 26. Stress relaxation modulus G(t) of NB70-PCS2, fit FIG. 28. The complex shear modulus G (ω) = G (ω)+iG (ω) of NB70-PCS2, fit by the FM model’s complex shear modulus by the FM model’s stress relaxation modulus Gfm(t). ∗ Gfm(ω).

30 η(t) icones (partial crosslinking, adding TFS, and adding sil- 20 ηFM (t) γ· = 0.0001 s -1 icone gum) are combined in a single non-simple borosil- G = 0.269 MPa (MPa·s) NB70-PCS2 icone, the observed β value is somewhat smaller than if 10 β = 0.807 η(t) one of those approaches had been omitted. τ = 44 s What is clear, however, is that preparing liquid borosil- 0 30 icones with β < 0.8 is difficult. Smaller β values require greater permanent-crosslink concentrations and can eas- 20 ily lead to gelation. A more promising path to small β γ· = 0.001 s -1 G = 0.251 MPa values is to stop trying to avoid gelation. Instead, cross (MPa·s) 10 β = 0.832 the gelation threshold deliberately and thereby produce η(t) τ = 54 s solid borosilicones. 0 0 500 1000 1500 Time (s) VISCOELASTIC SILICONE RUBBER FIG. 27. Shear viscosity η(t) of NB70-PCS2 at two different shear ratesγ ˙ . Each measurement is fit by the FM model’s Preparing partially-crosslinked silicone fluids close to shear viscosity ηfm(t). the gelation threshold is difficult and most attempts re- sulted in sticky, gelatinous goos. PCS2 was made in the attempt that came closest to the threshold without ac- had become so viscous that gelation was at most min- tually it. utes away, it was cooled to room temperature to stop the Once a partially-crosslinked silicone has crossed the crosslinking reaction. threshold and gelled, turning it into a homogeneous The viscosity of this PCS2 was 26.3 Pa·s, approxi- borosilicone is nearly impossible. To produce borosili- mately 400 times the viscosity of the original fluid (SFR cones beyond the gelation threshold, the TMB must be 70). PCS2 was then converted into a borosilicone by added while the partial-crosslinking is still in progress adding 1.50 wt% TMB and vacuum drying. The resulting and before gelation has occurred. Even that approach NB70-PCS2 was pressed into a rough sheet and allowed has difficulties because TMB interferes with the partial- to equilibrate with laboratory air for 10 days. crosslinking process and stiffens the still-crosslinking ma- The properties of the dry NB70-PCS2 were then mea- terial. sured and fit to the FM model, with the primary goal of Despite those complications, partial-crosslinking can determining the fractional order β. The stress relaxation produce borosilicones beyond the gelation threshold[4]. modulus G(t) (Fig. 26) was measured repeatedly and fit With their material-spanning permanent networks, these by Gfm(t) to obtain β = 0.785 ± 0.054. The shear vis- solid borosilicones cannot flow in response to stress and cosity η(t) (Fig. 27) was measured at two constant shear will be referred to as borosilicone rubbers or viscoelastic ratesγ ˙ and fit by ηfm(t) to obtain β = 0.820 ± 0.025. silicone rubbers (VSRs). And the complex shear modulus G∗(ω) (Fig. 28) was To produce a VSR that is free of behavior-influencing ∗ measured and fit by Gfm(ω) to obtain β = 0.790. catalysts, the partial crosslinking was done using VTAS Based on these values, β at the gelation threshold is and high temperatures. The starting fluid for this prepa- estimated as 0.80 ± 0.03. That threshold value, how- ration was PCS1, the partially-crosslinked fluid used to ever, probably depends somewhat on the detailed net- make NB70-PCS1. Since PCS1 is already close to the work structure of borosilicone, particularly the function- gelation threshold, it can be driven across that threshold ality and coordination of the crosslinkers. Moreover, by a small increase in the concentration of permanent when the three approaches to making non-simple borosil- crosslinks. 22

1 0.9 0.8 VSR70-PCS G(t) Spring S 0.7 E = 0.412 MPa G (t) FM G = 0.508 MPa Spring E (MPa) 0.6 β = 0.699 β Spring-pot SP G(t) 0.5 τ = 25.8 s 0.4 10 -2 10 -1 10 0 10 1 10 2 10 3 10 4 10 5 FIG. 29. The Fractional Zener Model consists of a spring E Time (s) in parallel with the Fractional Maxwell model, which itself consists of spring S in series with spring-pot SP . FIG. 30. Stress relaxation modulus G(t) of VSR70-PCS, fit by the FZ model’s stress relaxation modulus Gzm(t). Elastic modulus E, modulus G, fractional order β, and characteristic time τ are the fit’s four parameters. To 15g of PCS1 were added 4.5 wt% VTAS and 1.11 wt% TMB, followed immediately by 5 seconds of vigorous ∗ stirring and 30 seconds of vacuum drying. The mixture modulus Gfz(ω) is contained enough permanent crosslinker (VTAS) to well- ∗ ∗ ∗ exceed the gelation threshold. It also contained enough Gfz(ω) = Ge(ω) + Gfm(ω) temporary crosslinker (TMB) to become a borosilicone. (iωτ)β = E + G (74) It began thickening immediately. 1 + (iωτ)β Before the mixture could gel, it was transferred into two aluminum disk molds (38 mm dia x 6.35 mm) and its stress relaxation modulus and heated under pressure to 200 ◦C for five minutes. G (t) = G (t) + G (t) The high temperature greatly increased the permanent fz e fm β
crosslinking rate, allowing the material to solidify into = E + GEβ −(t/τ) (75) disks of viscoelastic silicone rubber, designated VSR70- It will be shown that the FZ model’s predictions fit PCS. the measured behaviors of viscoelastic silicone rubbers To prevent bubble formation, the VSR70-PCS disks remarkably well. While there are other fractional-order were cooled under pressure to room temperature. Once viscoelastic models for solids, they are not needed here. removed from their molds, the disks were transparent and bubble-free, but they contained methanol and acetic acid derived from the VTAS and TMB. To eliminate those Stress Relaxation Modulus volatile compounds and their esters, the disks were dried ◦ in 70 C air for several weeks. The stress relaxation modulus G(t) for VSR70-PCS was measured using the sudden compression technique. One such measurement is shown in Fig. 30, along with a The Fractional Zener Model fit by Gfz. The fit is excellent over 7 orders of magnitude in time, demonstrating how well the FZ model describes the stress-strain behavior of VSR70-PCS under compres- VSR70-PCS is a solid with viscoelastic properties like sion. a non-simple borosilicone. To describe its behavior, a vis- The fit to each measurement of G(t) provided values coelastic model must combine the fractional-order stress- for the model’s four parameters: the elastic modulus E strain response of a non-simple borosilicone for ω > 0 and associated with the spring component, the FM modu- t < ∞ and the stress-strain response of a solid at ω = 0 lus G associated with the FM component, the fractional and t → ∞. The simplest viscoelastic model that satis- order β, and the characteristic time τ. fies those requirements is the Fractional Zener model. VSR70-PCS’s elastic modulus is E = 0.391 ± 0.102 Shown in Fig. 29, the Fractional Zener model consists MPa. E represents the stiffness of VSR70-PCS’s perma- of two components in parallel: spring E and a Fractional nent network, present at all timescales and visible in Fig. Maxwell component. The Fractional Maxwell component 30 as VSR70-PCS’s modulus at large times. itself consists of two components in series: spring S and VSR70-PCS’s FM modulus is G = 0.539 ± 0.072 MPa. spring-pot SP . G reflects the added stiffness present immediately after The properties of the Fractional Zener (FZ) model can compression due to the temporary network piggybacking be calculated from those of its two component. Spring on the permanent network, the network liquid piggyback- ∗ E has Ge(ω) = E and Ge(t) = E. The FM component ing on the network solid. It is visible in Fig. 30 as the ∗ has Gfm(ω) given in Eq. (51) and Gfm(t) given in Eq. difference between VSR70-PCS’s modulus at small times (55). Using the parallel formula, the FZ model’s complex and its modulus at large times. 23

10 0 VSR70-PCS TABLE III. Compositions of four VSRs produced using the E = 0.60 au addition reaction. 0.225g Wacker Batch Aux 2 Pt catalyst was (au) G = 1.29 au added to each composition before vacuum drying and curing 10 -1 G'(ω) G''(ω) β = 0.68 at 200 ◦C for 5 minutes.

G*(ω) τ = 11 s G' FM (ω) G''FM (ω) 10 -2 Name PCS1 (g) TMB (g) SF 201 (g) XL-1 (g) -3 -2 -1 0 1 2 3 VSR70-A1 15.000 0.225 0.000 0.375 10 10 10 10 10 10 10 VSR70-A2 11.250 0.169 3.750 0.337 ω (1/s) VSR70-A3 7.500 0.112 7.500 0.300 FIG. 31. The complex shear modulus G∗(ω) = G0(ω)+iG00(ω) of VSR70-PCS, fit by the FZ model’s complex shear modulus ∗ Gfm(ω). the non-simple borosilicone’s subnetworks. When a sil- icone fluid containing hydride groups (Si-H) is added to that borosilicone, hydride groups can undergo ad- Immediately after compression, the temporary net- dition reactions with vinyl groups, forming permanent work contributes its modulus G to the stiffness of VSR70- crosslinks (Si-CH -CH -Si) between the silicone chains. PCS as though its crosslinks were permanent. The in- 2 2 With the help of these hydride-vinyl crosslinks, it is pos- stantaneous modulus G(0) is thus the sum of the individ- sible to exceed the gelation threshold and transform the ual moduli G(0) = E+G. With increasing time, however, liquid borosilicones in a viscoelastic silicone rubber[6]. the temporary network relaxes and VSR70-PCS’s stiff- The addition reaction is easily controlled with cata- ness decreases to G(∞) = E. The instantaneous modulus lyst and temperature, making it possible to prepare and has a temporary fraction G/(E + G) = 0.58 that decays dry a liquid borosilicone containing vinyl and hydride with time and a permanent fraction E/(E + G) = 0.42 group and then cure that liquid into a viscoelastic sili- that remains indefinitely. cone rubber. With a platinum catalyst and elevated tem- VSR70-PCS’s fractional order is β = 0.679±0.024, sig- peratures, the curing process can take minutes or less. nificantly smaller than that observed in the liquid borosil- Although the platinum catalyst remains in the finished icones. Although the RA model does not predict gela- VSR, it is present in such minuscule concentration that tion, it does predict a decreasing β as the volume con- it has no observable effects on the resulting VSR. centration of viscous cuboids decreases and that of elas- tic cuboids increases. Compared to any of the liquid Since PCS1 was partially crosslinked using VTAS, it borosilicones, the concentration of temporary crosslinks contains a high concentration of vinyl groups and can be in VSR70-PCS is smaller and the concentration of per- cured into VSR via the addition reaction. Three different manent crosslinks is larger, so VSR70-PCS’s smaller β is VSRs (Table III) were prepared by combining PCS1 with consistent with the RA model’s predictions. TMB, a hydride-containing polysiloxane, and a platinum catalyst, followed by drying and heating to 200 ◦C for 5 minutes. The hydride-containing polysiloxane was Masil Complex Shear Modulus XL-1 (Dystar), a copolymer with approximately 25% hy- dromethylsiloxane monomers. The platinum catalyst was A measurement of the complex shear modulus for Wacker Batch Aux 2 Pt. VSR70-PCS is shown in Fig. 31, along with a fit to In two of the VSRs, some of the borosilicone was re- ∗ placed by 1000 cSt vinyl-terminated PDMS (Masil SF Gfz(ω). The fit gives the fractional order β = 0.678 ± 0.017, consistent with the value obtained from the G(t) 201, Dystar). This substitution diluted the borosilicone measurement. and lowered the concentration of temporary crosslinks. Although the elastic modulus E and FM modulus G VSR70-A1 is undiluted while VSR70-A2 is 75% borosil- are reported in arbitrary units, the instantaneous modu- icone and VSR70-A3 is 50% borosilicone. lus has temporary fraction G/(E +G) = 0.68 and perma- Figure 32 shows the stress relaxation modulus G(t) nent fraction E/(E +G) = 0.32. Those two values differs measured for VSR70-A1 through VSR70-A3, along with somewhat from the values obtained from G(t). The ori- fits by Gfz(t). Those fits yield values for the elastic mod- gin of that difference is not yet understood. ulus E, FM modulus G, fractional order β, and charac- teristic time τ. Many such measurements were made and the values obtained are summarized in Table IV. τ, which Other Viscoelastic Silicone Rubbers depends sensitively on chemical environment, is omitted. As the borosilicone fraction decreases from 100% in To prepare VSRs with even higher permanent-crosslink VSR70-A1, to 75% in VSR70-A2, to 50% in VSR70-A3, concentrations, a second crosslinking technique was used. the concentration of temporary crosslinks decreases and In addition to forming permanent crosslinks, vinyltriace- so do the fractional order β, the FM modulus G, and toxylsilane incorporates vinyl groups (Si-CH=CH2) into the elastic modulus E. The decrease in β is consistent 24

1 10 0 VSR70-A1 VSR70-A1 G(t) G'(ω) G''(ω) G (t) E = 0.055 MPa FM G' FM (ω) G''FM (ω) 0.3 -1 E = 0.105 au (au) 10 (MPa) G = 1.108 MPa G = 1.600 au 0.1 β = 0.763

G(t) β = 0.779 -2 τ = 57.5 s G*(ω) 10 τ = 55 s 0.03 1 10 0 VSR70-A2 VSR70-A2 0.3 E = 0.026 MPa -1 E = 0.18 au (MPa) (au) 10 0.1 G = 0.751 MPa G = 1.57 au

G(t) β = 0.660 β = 0.74 -2

τ = 75.3 s G*(ω) 0.03 10 τ = 40 s

10 0 VSR70-A3 0.3 VSR70-A3 E = 0.013 MPa 0.1 -1 E = 0.16 au (au) 10 G = 0.482 MPa G = 1.31 au (MPa) β = 0.541 β = 0.63 0.03 -2 G(t) τ = 116.5 s G*(ω) 10 τ = 57 s 0.01 -4 -3 -2 -1 0 1 2 3 10 -2 10 -1 10 0 10 1 10 2 10 3 10 4 10 5 10 10 10 10 10 10 10 10 Time (s) ω (1/s)

FIG. 32. Stress relaxation modulus G(t) of VSR70-A1 FIG. 33. The complex shear modulus G∗(ω) = G0(ω)+iG00(ω) through A3, fit by the FZ model’s stress relaxation modu- of VSR70-A1 through A3, fit by the FZ model’s complex shear ∗ lus Gfz(t). Elastic modulus E, modulus G, fractional order modulus Gfz(ω). Elastic modulus E, modulus G, fractional β, and characteristic time τ are the fit’s four parameters. order β, and characteristic time τ are the fit’s four parameters.

TABLE IV. Elastic modulus E, FM modulus G, and frac- TABLE V. Elastic modulus fraction E/(E + G), FM modu- tional order β obtained by fitting the FZ model’s stress re- lus fraction G/(E + G), and fractional-order β obtained by ∗ laxation modulus Gfz(t) to stress relaxation modulus G(t) for fitting the FZ model’s complex modulus Gfz(ω) to complex multiple measurements of VSR70-A1 through VSR70-A3. shear modulus G∗(ω) for multiple measurements of VSR70- A1 through VSR70-A3. Name E (MPa) G (MPa) β VSR70-A1 0.061 ± 0.009 1.065 ± 0.200 0.777 ± 0.031 Name E/(E+G) G/(E+G) β VSR70-A2 0.030 ± 0.013 0.721 ± 0.116 0.677 ± 0.067 VSR70-A1 0.062 ± 0.003 0.938 ± 0.003 0.777 ± 0.007 VSR70-A3 0.024 ± 0.023 0.397 ± 0.092 0.570 ± 0.132 VSR70-A2 0.086 ± 0.005 0.914 ± 0.005 0.732 ± 0.012 VSR70-A3 0.132 ± 0.008 0.868 ± 0.008 0.631 ± 0.015 with the RA model, which predicts such a decrease as the fractional concentration of viscous cuboids decreases. tic modulus fraction E/(E + G), FM modulus fraction G is the modulus of the temporary network, the net- G/(E + G), and fractional-order β obtained from multi- work liquid piggybacking on the network solid. That ple measurements are shown in Table V. G decreases almost in proportion to VSR’s borosilicone For VSR70-A1 (100% borosilicone), the fractional or- fraction indicates that G is due primarily to that borosil- der β and the modulus fractions E/(E + G) and G/(E + icone fraction. That is expected since, without the G) agree well with values obtained from the stress relax- borosilicone and its temporary crosslinks, there would ation modulus G(t). be no Fractional Maxwell behavior. For VSR70-A2 (75% borosilicone) and VSR70-A3 E is the modulus of the permanent network, the net- (50% borosilicone), β agrees with the value obtained from work solid. In these VSRs, E decreases as the borosili- the stress relaxation modulus G(t), but modulus frac- cone fraction decreases because SF 201 forms a softer per- tions do not. In both cases, the complex shear modulus manent network that does PCS1. The average molecule measurements observe a somewhat larger elastic modu- in PCS1 has multiple vinyl groups linked by short PDMS lus fraction than is observed in the stress relaxation mea- chains, whereas an SF 201 molecule has two vinyl groups surements. That disagreement is probably due in part to linked by a long PDMS chain. Reducing the borosili- the presence of two very different PDMS chain lengths in cone fraction thus has the side effect of weakening the these diluted-borosilicone VSRs. permanent network and decreasing E. Beyond basic science, viscoelastic silicone rubbers ex- Figure 33 shows the complex shear modulus G∗(ω) for hibit shape memory behaviors that make them useful in ∗ those same three VSRs, along with fits by Gfz(ω). Elas- practical applications. Because a VSR’s network liquid 25 component opposes sudden changes in shape, it takes (2.45 wt% TMB in SFR 70) were prepared. One ver- time for the VSR to “learn” a new shape or “forget” sion was simply vacuum-dried, the second version was a previous shape. When subject to external-imposed vacuum-dried and then allowed to absorb ∼110 ppm change in strain, the VSR responds with a large initial (∼0.011 wt%) water from laboratory air, and the third stress that gradually relaxes as the VSR complies with version had 4 ppm (0.0004 wt%) iso-stearic acid (ISA) the new strain. When left alone, the VSR’s network solid added before vacuum-drying. The vacuum-dried version component gradually returns the VSR to the shape in had τ = 40.2 s, the 110 ppm water version had τ = 8.0 which it was originally formed. s, and the 4 ppm ISA version had τ = 13.6 s. It follows that ISA molecules reduce τtc approximately 250 times as effectively as water molecules. ABOUT THE TEMPORARY CROSSLINKS Alcohols also reduce τ, though less effectively than carboxylic acids or water. Moreover, alcohols tend to Each model used to describe a borosilicone or VSR weaken or even liquefy borosilicones, an effect that makes has a characteristic time τ. The accuracy with which the alcohols useful for cleaning purposes. It may be that al- models describe the behaviors of the materials suggests cohol molecules substitute rapidly for PDMS chains on that τ is closely related to some physical process in the boron, but that the reverse substitution is considerably materials themselves. slower. If the resulting alkoxy groups on boron resist sub- In the TN model, from which the other models even- stitution, τ will decrease only slightly and the transient tually emerged, τ is the mean lifetime of a strand. Given network will be weakened or even discontinuous. that temporary crosslinks emulate strand-breaking, it is Silanol groups themselves do not reduce τ significantly, reasonable to suppose that τ arises from and is propor- probably because silanol-terminated PDMS chains rarely tional to the mean lifetime of a temporary crosslink τtc, substitute for one another directly on boron crosslinks. As evidence for this limited exchange, consider SB40-

τ = abτtc, (76) 4.50, a borosilicone with only 55% of the boron crosslinks needed for stoichometric saturation (Table I). Despite be- where ab may depend on the network structure of the ing at least 1.80 wt% OH in the form of terminal silanol borosilicone or VSR. Given that simple borosilicones groups on PDMS chains, its τ can exceed 10 s. It follows have only temporary network structure, it is likely that that water molecules reduce τtc approximately 150 times ab ≈ 1. Because of their permanent subnetworks and as effectively as terminal silanol groups on PDMS. networks, non-simple borosilicones and VSRs have more The temperature dependence of a borosilicone’s com- ∗ difficulty relaxing stress and probably have ab > 1. plex shear modulus G (ω) provides insight into its dy- When a simple borosilicone has been cleanly prepared namics and energy barriers, so G∗(ω) was measured at and vacuum-dried, τ can be 40 s or more. When that several temperatures between 20 and 80 ◦C. These tem- same borosilicone is allowed to equilibrate with labora- perature studies were conducted on two of the SB70-2.45 tory air (20 ◦C, 50% relative humidity), τ is typically 15 versions: the version containing ∼110ppm water and the s or less. Since the water concentration in PDMS rub- version containing 4ppm ISA. ◦ ∗ ∗ ber equilibrated with 20 C, 50 %RH air is only ∼110 The measured G (ω) curves were well-fit by Gfm(ω) ppm (∼0.011 wt%),[3] water evidently reduces the mean (Figs. 34a,b and 35a,b), yielding β values just slightly lifetime of the temporary crosslinks catalytically. less than 1 and τ values that decrease modestly with in- The catalysis mechanism probably involves exchange creasing temperature. Arrhenius plots of ln(1/τ) vs 1/T reactions, with water molecules and silanol-terminate for both borosilicones are shown in Fig. 36. Assuming PDMS chains substituting for one another on boron Eq. (76), plots of ln(1/τtc) vs 1/T would look identical, crosslinks (Fig. 1, where R = H). In the first step, a free except for a ln(1/ab) vertical shift. water molecule substitutes for a PDMS chain on boron, The Arrhenius plot slopes for the 110ppm water and resulting in an OH group on boron and a free silanol- 4ppm ISA borosilicones give reaction energy barriers Ea terminated chain. In the second step, a free silanol- of 0.177 ± 0.039 eV (4.07 ± 0.90 kcal/M) and 0.181 ± terminated PDMS chain substitutes for the OH group 0.024 eV (4.16 ± 0.54 kcal/M), respectively. There is on boron, resulting in the PDMS chain on boron and no significant difference between those energy barriers, a free water molecule. This water-catalyzed breaking so the difference in catalytic activities must be due to and reforming of temporary crosslinks is analogous to reaction cross sections: ISA has a much larger reaction the strand breaking and reforming of the TN model. cross section than water. Carboxylic acids are far more effective than water at A material is said to be thermo-rheologically simple if, reducing τ, probably because -COOH groups offer more in the study of its linear viscoelastic behavior, a “change opportunities for successful exchanges than -OH groups. in temperature is equivalent to a shift of the logarithmic To compare the catalytic effectiveness of water and time scale”[38]. Figure 34 shows G∗(ω) measurements carboxylic acid, three versions of SB70-2.45 borosilicone of SB70-2.45 with ∼110ppm water taken at (a) 22.6 ◦C 26

SB70-2.45 with 110 ppm water -0.5 0 10 (a) -1 Water: E = 0.177 ± 0.039 eV G'(ω) G''(ω) a -1 22.6 °C G' (ω) G'' (ω) -1.5 (au) 10 FM FM (1/τ)

G = 0.93 au ln -2 -2 10 β = 0.96 -2.5 G*(ω) ISA: E = 0.181 ± 0.024 eV τ = 7.958 ± 0.153 s a -3 -3 10 2.8 2.9 3 3.1 3.2 3.3 3.4 3.5 -1 -3 0 (b) 10 1/T (K ) 10

-1 77.0 °C (au) 10 G = 3.05 au FIG. 36. Arrhenium plots for SB70-2.45 containing 110ppm β = 0.96 water (◦) or 4ppm ISA ( ). 10 -2 

G*(ω) τ = 2.626 ± 0.104 s 10 -3 10 0 and (b) 77.0 ◦C. Both measurements appear in (c), with (c) G 22.6 °C '(ω) ◦ -1 G'(ω) 77.0 °C the 77.0 C scaled vertically by the ratio of G values (au) 10 (×0.3050) and scaled horizontally by the ratio of τ val- 10 -2 G''(ω) 22.6 °C ues (×0.3300). Vertical scaling was necessary because of G*(ω) G''(ω) 77.0 °C changes in sample shape between the two measurements. 10 -3 -3 -2 -1 0 1 2 3 The excellent overlap of the two traces indicates that this 10 10 10 10 10 10 10 borosilicone exhibits thermo-rheological simplicity. In- ω (1/s) creasing the temperature of the borosilicone reduces its characteristic time τ and thereby shifts the entire G∗(ω) FIG. 34. The complex shear modulus G∗(ω) for SB70-2.45 with 110ppm water. (a) was measured at 22.6 ◦C and (b) at to higher angular frequencies ω. ∗ 77.0 ◦C. (c) combines both sets of data, but with the higher Figure 35 shows G (ω) measurements of SB70-2.45 temperature data set scaled vertically by the ratio of G values with 4ppm ISA taken at (a) 21.0 ◦C and (b) 80.0 ◦C. (×0.3050) and horizontally by the ratio of τ values (×0.3300). Again, both measurements appear in (c), with the 80.0 ◦C scaled vertically by ratio of G values (×0.8074) and scaled horizontally by the ratio of τ values (×0.2969). This borosilicone also exhibits thermo-rheological sim- SB70-2.45 with 4ppm ISA plicity. 0 10 G'(ω) G''(ω) G' (ω) G'' (ω) -1 FM FM (au) 10 21.0 °C CONCLUSIONS G = 1.65 au -2 β = 0.963

G*(ω) 10 τ = 13.573 ± 0.303 s While boron atoms act as trifunctional crosslinkers -3 10 that attach covalently to the ends of silicone polymer 10 0 chains, the simple borosilicones they produce are not

-1 80.0 °C solids. Instead of gelling when its boron concentration ex- (au) 10 G = 2.04 au ceeds the predicted gelation threshold, a simple borosili- 10 -2 β = 0.96 cone exhibits only a dramatic increase in shear viscosity. G*(ω) τ = 4.030 ± 0.112 s -3 Its viscosity continues to increase as more boron is added, 10 becoming extraordinarily large as the boron concentra- 0 10 (c) G'(ω) 21.0 °C tion approaches stoichiometric saturation. Regardless of G'(ω) 80.0 °C -1 boron concentration, a simple borosilicone never solidi- (au) 10 fies. The only plausible explanation for such observations 10 -2 G''(ω) 21.0 °C G*(ω) G''(ω) 80.0 °C is that boron crosslinks are temporary. 10 -3 Boron crosslinks are temporary not because they are 10 -3 10 -2 10 -1 10 0 10 1 10 2 10 3 weak, but because they undergo easy substitution. The ω (1/s) presence of -OH bearing molecules in the borosilicone al- lows for exchange reactions in which one -OH bearing FIG. 35. The complex shear modulus G∗(ω) for SB70-2.45 molecular substitutes for another -OH bearing molecule with 4ppm ISA. (a) was measured at 21.0 ◦C and (b) at 80.0 ◦ on a boron crosslink. This exchange process is partic- C. (c) combines both sets of data, but with the higher tem- ularly efficient between carboxylic acids (R-COOH) and perature data set was normalized vertically (×0.8074) and shifted horizontally by the ratio of the τ values (×0.2969). silanol-terminated silicones (OH-PDMS-OH). The relent- less substitution of one attachment for another on each boron-silicone crosslink gives those crosslinks a mean life- 27 time that is typically measured in seconds or less. On the shortest timescales and at the highest frequen- Because boron crosslinks are temporary, borosilicones cies, the liquid network has no time to evolve and its full are network liquids. Simple borosilicones, in which the modulus is also observed. At intermediate timescales and silicone polymer chains are coupled only by temporary frequencies, the dynamics of the liquid network are im- boron crosslinks, are excellent physical realizations of portant and the VSR exhibits behaviors described by the Green and Tobolsky’s transient network model. Apart fractional-order differential equations of the FM model. from some excess loss at large ω, the observed moduli and The fractional order β depends primarily on the concen- viscosities of these simple borosilicones agree well with tration of temporary crosslinks in the VSR and decreases the predictions of the TN model and its Lodge Elastic as that concentration decreases. Fluid. The Maxwell model, which defines a simple elas- In the end, borosilicones and VSRs are remarkably sim- tic fluid in terms of integer-order viscoelastic elements, ple. On a molecular level, they are almost Tinkertoy R shares those same predictions. systems: vast assemblies of elastic strands joined by When permanent crosslinks are present in a borosili- temporary and permanent links or hubs. Their struc- cone, it no longer behaves as a simple elastic fluid. Its tural randomness and the statistical breaking of their time-dependences are slower than exponential and its temporary crosslinks further contribute to their simplic- storage and loss moduli decrease more slowly than ω2 and ity. They permit simple structural modeling, including ω, respectively, as ω → 0. Such observations cannot be the transient network and random assembly models, and modeled with finite assemblies of integer-order viscoelas- their behaviors can are beautifully characterized by few- tic elements and the associated integer-order differential element viscoelastic models, including the Maxwell, Frac- equations, so another approach is required. tional Maxwell, and Fractional Zener models. The Random Assembly model, composed of infinites- These materials are not absolutely perfect, of course, imal viscous and elastic bodies representing transient so measurement and theory sometimes differ. What is and non-transient couplings, provides a path forward. consistent throughout this work, however, is the dom- This model leads analytically and computationally to the inant role a single characteristic time τ plays in every spring-pot, a fractional-order viscoelastic element defined measurement and every model of a borosilicone or VSR. by fractional-order differential equations. The spring- In each case, τ derives from the mean lifetime of the ma- pot’s fractional order is β, which is also the volume con- terial’s temporary crosslinks and is the only significant centration of viscous cuboids in the Random Assembly timescale observed in the material. model. The reciprocal of τ follows the Arrhenius equation, When a spring is put in series with the spring-pot, to so the effect of a change in temperature is a change in eliminate the spring-pot’s divergence at ω → ∞ and rec- τ and therefore a change in the material’s only signif- ognize the finite compressibility of materials, the result icant timescale. This behavior, in which a change in is the Fractional Maxwell viscoelastic model. The ob- temperature causes only a change in timescale, is known served moduli and viscosities of non-simple borosilicones as thermo-rheological simplicity. The borosilicones and agree well with the predictions of the FM model. The VSRs studied in this work are thermo-rheologically sim- slower-than-exponential time-dependences observed ex- ple. perimentally correspond to Mittag-Leffler functions ob- tained from the FM model’s fractional-order differential equations and the slow variations in storage and loss moduli observed as ω → 0 correspond to ωβ terms in the FM model solutions. ∗ [email protected] When the concentration of permanent crosslinks [1] EW Abel and Apar Singh. The trialkylsilyl esters of in a borosilicone exceeds the gelation threshold, its boron. Journal of the Chemical Society (Resumed), pages permanently-crosslinked subnetworks coalesce into a 690–693, 1959. material-spanning network and the borosilicone be- [2] DP Almond and B Vainas. The dielectric properties of random r-c networks as an explanation of theuniver- comes viscoelastic silicone rubber. The permanently- sal’power law dielectric response of solids. Journal of crosslinked network renders the VSR a network solid and Physics: Condensed Matter, 11(46):9081, 1999. gives it an equilibrium shape to which it returns when free [3] JA Barrie and D Machin. The sorption and diffusion of of external influences. The VSR’s temporary crosslinks, water in silicone rubbers: Part i. unfilled rubbers. Jour- however, still form a liquid network, one so closely cou- nal of Macromolecular Science, Part B: Physics, 3(4): pled to the solid network solid that it effectively “piggy- 645–672, 1969. backs” on the solid network. [4] Louis A. Bloomfield. U.S. Patent No. 9,102,799, 11 Au- gust 2015. The measured moduli of VSRs are well-described by [5] Louis A Bloomfield. A simple karl fischer titration tech- the Fractional Zener model: a spring in parallel with the nique for measuring the silanol content of high molecular FM model. On the longest timescales and at the lowest weight polysiloxanes. International Journal of Polymer frequencies, only the solid network’s modulus is observed. Analysis and Characterization, 17(5):394–399, 2012. 28

[6] Louis A. Bloomfield. U.S. Patent No. 9,550,864, 24 Jan- [29] Siegfried Nitzsche and Manfred H. Wick. U.S. Patent No. uary 2017. 2,842,521, 8 July 1958. [7] Reginald J. Boot. U.S. Patent No. 3,213,048, 19 Octo- [30] Walter Noll. Chemistry and Technology of Silicones. Aca- ber 1965. demic Press, 1968. [8] CR Bowen and Darryl P Almond. Modelling [31] Note1. While transient couplings might be expected to the’universal’dielectric response in heterogeneous mate- produce elastic fluid behavior rather than viscous behav- rials using microstructural electrical networks. Materials ior, models based on that choice do not predict the be- Science and Technology, 22(6):719–724, 2006. havior of non-simple borosilicones. [9] Rod Cross. Elastic and viscous properties of silly putty. [32] Richard G Palmer, Daniel L Stein, Elihu Abrahams, and American Journal of Physics, 80(10):870–875, 2012. Philip Warren Anderson. Models of hierarchically con- [10] Edmundo Capelas De Oliveira, Tenreiro Machado JA, strained dynamics for glassy relaxation. Physical Review et al. A review of definitions for fractional derivatives and Letters, 53(10):958, 1984. integral. Mathematical Problems in Engineering, 2014. [33] Igor Podlubny. Fractional differential equations: an in- [11] Norbert G. Dickmann. U.S. Patent No. 2,721,857, 25 Oc- troduction to fractional derivatives, fractional differential tober 1955. equations, to methods of their solution and some of their [12] Paul J Flory. Molecular size distribution in three dimen- applications, volume 198. Academic press, 1998. sional polymers. i. gelation. J. Am. Chem. Soc, 63(11): [34] Eugene G. Rochow. U.S. Patent No. 2,371,068, 3083–3090, 1941. 6 March 1945. [13] Paul J Flory. Molecular size distribution in three dimen- [35] Prince E Rouse Jr. A theory of the linear viscoelastic sional polymers. ii. trifunctional branching units. J. Am. properties of dilute solutions of coiling polymers. The Chem. Soc, 63(11):3091–3096, 1941. Journal of Chemical Physics, 21(7):1272–1280, 1953. [14] George Grosvenor Freeman. Silicones. ILIFFE, 1962. [36] H Schiessel and A Blumen. Hierarchical analogues to [15] MS Green and AV Tobolsky. A new approach to the the- fractional relaxation equations. Journal of Physics A: ory of relaxing polymeric media. The Journal of Chemical Mathematical and General, 26(19):5057, 1993. Physics, 14(2):80–92, 1946. [37] Helmut Schiessel, Alexander Blumen, et al. Mesoscopic [16] Hans J Haubold, Arak M Mathai, and Ram K Saxena. pictures of the sol-gel transition: ladder models and frac- Mittag-leffler functions and their applications. Journal tal networks. Macromolecules, 28(11):4013–4019, 1995. of Applied Mathematics, 2011, 2011. [38] F Schwarzl and AJ Staverman. Time-temperature depen- [17] R Heyes and JC Lockhart. Redistribution reactions on dence of linear viscoelastic behavior. Journal of Applied boron. exchange of alkoxy-groups on boric and phenyl- Physics, 23(8):838–843, 1952. boronic esters. Journal of the Chemical Society A: Inor- [39] Montgomery T Shaw. Introduction to polymer rheology. ganic, Physical, Theoretical, pages 326–328, 1968. John Wiley & Sons, 2012. [18] Nicole Heymans and J-C Bauwens. Fractal rheological [40] MT Shaw and WJ MacKnight. Introduction to Polymer models and fractional differential equations for viscoelas- Viscoelasticity. Wiley-Interscience, 2005. tic behavior. Rheologica acta, 33(3):210–219, 1994. [41] Leslie H Sperling. Introduction to physical polymer sci- [19] RC Koeller. Applications of fractional calculus to the ence. John Wiley & Sons, 2006. theory of viscoelasticity. ASME, Transactions, Journal of [42] MD Stern and AV Tobolsky. Stress-time-temperature Applied Mechanics(ISSN 0021-8936), 51:299–307, 1984. relations in polysulfide rubbers. The Journal of Chemical [20] Rudolph Kohlrausch. Theorie des elektrischen Physics, 14(2):93–100, 1946. r¨uckstandes in der leidener flasche. Annalen der [43] Walter H Stockmayer. Theory of molecular size distribu- Physik, 167(2):179–214, 1854. tion and gel formation in branched polymers ii. general [21] Roderic S Lakes. Viscoelastic materials. Cambridge Uni- cross linking. The Journal of Chemical Physics, 12(4): versity Press, 2009. 125–131, 1944. [22] RG Larson. Constitutive Equations for Polymer Melts [44] Van-Tan Truong and John G Ternan. Complex conduc- and Solutions. Butterworths, 1988. tivity of a conducting polymer composite at microwave [23] Karl Lichtenecker. Die dielektrizit¨atskonstante frequencies. Polymer, 36(5):905–909, 1995. nat¨urlicher und k¨unstlicher mischk¨orper. Physikaliche [45] M Wick. Bor-siloxan-elastomere. Kunststoffe, 50:433– Zeitschrift, 27:115–158, 1926. 436, 1960. [24] AS Lodge. A network theory of flow birefringence and [46] JG Williams and C Gamonpilas. Using the simple com- stress in concentrated polymer solutions. Transactions of pression test to determine youngs modulus, poissons ratio the Faraday Society, 52:120–130, 1956. and the coulomb friction coefficient. International Jour- [25] JE Mark, RR Rahalkar, and JL Sullivan. Model net- nal of Solids and Structures, 45(16):4448–4459, 2008. works of end-linked polydimethylsiloxane chains. iii. ef- [47] James G. E. Wright. U.S. Patent No. 2,541,851, 13 Febru- fect of the functionality of the cross-links. The Journal ary 1951. of Chemical Physics, 70(4):1794–1797, 1979. [48] Yugong Wu, Xuanhe Zhao, Fei Li, and Zhigang Fan. [26] AM Mathai. Mittag-leffler functions and fractional cal- Evaluation of mixing rules for dielectric constants of com- culus. Recent Developments and Recent Applications in posite dielectrics by mc-fem calculation on 3d cubic lat- Statistics and Astrophysics, 32:71, 2005. tice. Journal of Electroceramics, 11(3):227–239, 2003. [27] Rob Roy McGregor and Earl Leathen Warrick. U.S. [49] TI Zatsepina, ML Brodskii, Ye A Frolova, Patent No. 2,431,878, 2 December 1947. AA Trapeznikov, VN Gruber, and GA Kruglova. [28] LA Mitrofanov, Ye A Sidorovich, AV Karlin, and Rheological behaviour of polyheterosiloxanes. Polymer AI Marei. The question of molecular interaction in Science USSR, 12(11):2899–2905, 1970. polyborondimethylsiloxanes (pbdms). Polymer Science [50] Bruno H Zimm. Dynamics of polymer molecules in di- USSR, 11(4):882–888, 1969. lute solution: viscoelasticity, flow birefringence and di- 29 electric loss. The journal of chemical physics, 24(2):269– 278, 1956.