張 鵬
(1.長江師范學(xué)院 電子信息工程學(xué)院,重慶 408003;2.華中科技大學(xué) 同濟(jì)醫(yī)學(xué)院,湖北 武漢 430030)
為了簡單起見,本文只針對線性聚合物進(jìn)行討論。另外,值得討論的是,帶加寬效應(yīng)可以由色譜柱、內(nèi)部檢測器或者兩者共同引起。目前的方法僅使用單一探測器獲得的SEC數(shù)據(jù)。如果實(shí)驗(yàn)使用多探測器,那么該方法需要分別應(yīng)用于每個(gè)探測器。
帶加寬效應(yīng)的影響通常采用Tung卷積方程[15]來描述:
(1)
其中,V是洗脫體積,S(V)是SEC探測信號(hào)(實(shí)際測量的SEC分布曲線),w(logM)是準(zhǔn)確的SEC分布曲線(沒有帶加寬的曲線),M是分子量。G(logM,V)是加寬函數(shù)。我們選擇國際上廣泛被認(rèn)可的EGH函數(shù)作為加寬函數(shù):
(2)
這里σ和τ是加寬函數(shù)的兩個(gè)參數(shù),C是歸一化常數(shù)。
圖1展示了本文提出的方法思路:
1、將一組不同樣品的SEC觀測信號(hào)(也就是測量得到的SEC分布曲線,S(V))進(jìn)行退卷積處理(方程(1)的逆運(yùn)算)得到相應(yīng)的一組退卷積后的SEC分布曲線,w(logM)。在第一次退卷積過程中,任意賦予加寬函數(shù)中的兩個(gè)參數(shù)σ和τ各一個(gè)初始值。
2、分別對每一條退卷積后的SEC分布曲線,w(logM), 依據(jù)方程(3):
(3)
(4)
的值盡可能為零。這里,n是樣品的數(shù)目。此時(shí)σ和τ值就是我們最終所需要的結(jié)果。這樣,就可以運(yùn)用這一組σ和τ值退卷積任意一條由該SEC儀器得到的分布曲線(對于不同的儀器需要重新計(jì)算獲得一組σ和τ值)得到準(zhǔn)確的未加寬的分布曲線。
(6)
圖1 方法流程圖Fig.1 Illustrating the new method
在本文提出的方法中,正確的退卷積運(yùn)算對最終的結(jié)果起著非常重要的作用。因?yàn)殡m然理論上退卷積的曲線必然會(huì)和原來的實(shí)驗(yàn)重合,但是準(zhǔn)確的退卷積的過程依賴正確的運(yùn)算方法。不恰當(dāng)?shù)倪\(yùn)算操作可能會(huì)導(dǎo)致錯(cuò)誤的結(jié)果,所以在本文中有必要采用可靠的退卷積方法。圖2展示了用傳統(tǒng)Ishige退卷積算法和本文加入了最大熵算法(方程(6))的修正退卷積算法運(yùn)算的結(jié)果比較。顯然,圖1(a)中,如果實(shí)驗(yàn)SEC分布曲線用Ishige算法退卷積,結(jié)果會(huì)有很大的漲落,而且退卷積運(yùn)算的迭代次數(shù)越多,漲落越大。這里迭代次數(shù)是1000次。這是因?yàn)閷?shí)驗(yàn)的分布曲線中即使非常小的一點(diǎn)漲落會(huì)在迭代的過程中放大。圖2(b)展示了加入了最大熵算法的修正方法的退卷積結(jié)果,可以看出漲落消失了。
雖然,修正的退卷積方法可以消除漲落,但是為了進(jìn)一步證明修正的退卷積方法是正確的。我們進(jìn)行了測試。給定一系列σ 和 τ值利用方程(1)分別卷積一條實(shí)驗(yàn)SEC分布曲線。再用同樣的σ 和 τ值進(jìn)行退卷積運(yùn)算。如果退卷積的曲線可以和原來的實(shí)驗(yàn)曲線重合,那么說明修正的退卷積方法是正確的。圖3展示了測試的結(jié)果。
圖2 傳統(tǒng)Ishige退卷積算法和本文加入了最大熵算法的修正退卷積算法的運(yùn)算結(jié)果比較。(a)實(shí)驗(yàn)SEC數(shù)據(jù)用傳統(tǒng)Ishige退卷積算法結(jié)果;(b)實(shí)驗(yàn)SEC數(shù)據(jù)用本文加入了最大熵算法的退卷積算法(方程(6))的結(jié)果。其中σ=0.2以及τ=0.01。退卷積運(yùn)算的迭代次數(shù)是1000次。實(shí)驗(yàn)數(shù)據(jù)來自文獻(xiàn)[17]。所有數(shù)據(jù)均已歸一化。Fig.2 A comparison on results from traditional Ishige algorithm and that with maximum entropy method added upon for deconvolution procedure. (a) raw SEC data deconvoluted by traditional Ishige algorithm; (b) raw SEC data deconvoluted by Ishige algorithm with maximum entropy method. σ=0.2 and τ=0.01. Time of iteration in deconvolution is 1000 times. Raw SEC data was taken from [17]. All the data has been normalized.
圖3 (a)從上至下分別使用σ=0.2,τ=0.01;σ=0.4,τ=0.02以及σ=0.6,τ=0.03帶入方程(1)卷積的結(jié)果;(b)(a)圖的相應(yīng)退卷積結(jié)果。實(shí)驗(yàn)數(shù)據(jù)來自文獻(xiàn)[17]。所有數(shù)據(jù)均已歸一化。Fig.3 (a) Convoluted results respectively with given σ=0.2,τ=0.01;σ=0.4,τ=0.02 and σ=0.6,τ=0.03 (from top to bottom); (b) the corresponding deconvoluted results for (a) from top to bottom. Raw SEC data was taken from [17]. All the data has been normalized.
圖3(a)展示了給定的三組σ 和 τ值帶入方程(1)卷積一條原始實(shí)驗(yàn)曲線的結(jié)果。圖3(b)展示了相應(yīng)的退卷積結(jié)果與原始實(shí)驗(yàn)曲線的比較,與圖(a)相比,顯然它們重合的很好。證明修正的退卷積算法是正確的。圖4 (a)和(b)分別展示了在σ 和 τ空間中一條Metropolis抽樣軌跡以及方程(5)中χ隨計(jì)算迭代次數(shù)變化的曲線。本例中,σactual和τactual值分別是0.45和-0.02。σguess和τguess的值分別是0.70和0.00。顯然,軌跡成功地從從初始值(start)搜尋到目標(biāo)值(end)。相應(yīng)的,在迭代計(jì)算的過程中,χ也匯聚于穩(wěn)定值。
圖4 (a)在σ 和 τ空間中一條Metropolis抽樣軌跡;(b)方程(5)中χ隨計(jì)算迭代次數(shù)變化的曲線。Fig.4 (a) The Metropolis sampling trajectory of χ in space of σ and τ; (b) Tendency of value of χ with iteration time.
對于本文的方法,做了7組測試,如表1所示。為了定量的描述σconverged和τconverged的值與σactual和τactual的值的接近程度。本文給出了相對誤差δσ和 δτ。它們的表達(dá)式是
(7)
從表1中可以看出,σconverged和τconverged的值與σactual和τactual的值差別很小。但是為了討論這個(gè)差別是否會(huì)造成重大影響,本文用σconverged和τconverged以及σactual和τactual分別帶入方程(1)卷積同一系列原始SEC分布曲線(不同樣品的SEC分布曲線),比較卷積后的兩條曲線,若重合,則表示誤差造成的影響可以忽略。為了使結(jié)果有說服力,在此,本文選擇了表1中相對誤差最大的第一組σconverged和τconverged與σactual和τactual進(jìn)行計(jì)算。
表1 本文方法的7組測試,δσ 和 δτ 是相對誤差Table 1 Seven groups of test about our method, δσ and δτ are the relative error
圖5展示了比較的結(jié)果。曲線重合的很好,誤差不會(huì)造成影響。
圖5 (a)不同樣品的實(shí)驗(yàn)SEC分布曲線分別用σconverged和τconverged以及σactual和τactual卷積的結(jié)果比較。為了清楚展示,每條曲線的縱坐標(biāo)向上平移。Fig.5 A comparison on the convoluted results of raw SEC data using σconverged and τconvergedbetween using σactual and τactual. All the data has been normalized. In order to distinguish different samples, curves were shifted on the vertical axis. In here, the parameters in table 1 shows the largest relative error through seven groups of test were chosen, that are, σactualis 0.300, τactual is -0.030, σconverged is 0.309 and τconverged is -0.036.
針對由于帶加寬效應(yīng)的存在導(dǎo)致SEC分布曲線加寬、扭曲等現(xiàn)象,本文提出了一種有效的去除方法,并對此方法做出了詳盡的理論測試。測試的結(jié)果顯示從非SEC方法(沒有帶加寬效應(yīng))獲得的和可以被用來尋找SEC的加寬函數(shù)的參數(shù)。從而可以很好的通過退卷積運(yùn)算真實(shí)的SEC分布曲線。
[1] STRIEGEL A M, YAU W W, KIRKLAND J J, et al. Modern size-exclusion liquid chromatography-practice of gel permeation and gel filtration chromatography[M]. John Wiley & Sons, Hoboken, 2009.
[2] SCHN?LL-BITAI I, MADER C. How much does band broadening adulterate results deduced from chromatograms measured by size-exclusion chromatography really[J]. J Chromatog A, 2006(1137):198-206.
[3] CLAY P A, GILBERT R G. Molecular weight distributions in free-radical polymerizations. Model development and implications for data interpretation[J]. Macromolecules,1995(28):552-569.
[4] VAN BERKEL K Y, RUSSELL G T, GILBERT R G. Molecular weight distributions and chain-stopping events in the free-radical polymerization of methyl methacrylate[J]. Macromolecules,2005(38):3214-3224.
[5] GRUENDLING T, GUILHAUS M, BARNER-KOWOLLIK C. Fast and accurate determination of absolute individual molecular weight distributions from mixtures of polymers via size exclusion chromatography-electrospray ionization mass spectrometry[J]. Macromolecules, 2009(42):6366-6374.
[6] GRUENDLING T, VOLL D, GUILHAUS M, et al. A perfect couple: PLP/SEC/ESI-MS for the accurate determination of propagation rate coefficients in free radical polymerization[J]. Macromolecular Chemistry and Physics, 2010(211):80-90.
[7] SATO H, ICHIEDA N, TAO H, et al. Data processing method for the determination of accurate molecular weight distribution of polymers by SEC/MALDI-MS[J]. Analytical Sciences, 2004(20):1289-1294.
[8] GRITTI F, GUIOCHON G. The van deemter equation: assumptions, limits, and adjustment to modern high performance liquid chromatography[J]. Journal of Chromatography A, 2013(1302):1-13.
[9] MILLER C T, GRAY W G. On the algebraic and differential forms of Darcy's equation[J]. J Porous Media, 2011(14):33-50.
[10] BAUMGARTEN J L, BUSNEL J P, MEIRA G R. Band broadening in size exclusion chromatography of polymers. State of the art and some novel solutions[J]. J Liquid Chromatography & Related Technologies, 2002(25):1967-2001.
[11] SACTO P. A novel automatic dilution system for online particle size analysis[J]. ACS Symp Ser, 1998(693):23-29.
[12] YOSSEN M M. Determination of the band broadening function in size exclusion chromatography with light-scattering detection[J]. Journal of Liquid Chromatography & Related Technologies, 2012,35(1):79-94.
[13] GRAY J G, DUDAS S, CZUB S. A study on the analytical sensitivity of 6 BSE tests used by the canadian BSE reference laboratory[J]. PLoS One, 2011(6):e17633.
[14] GRUSHKA E. Characterisation of exponentially-modified gaussian peaks in chromatography[J]. Anal Chem, 1972(44):1733-1738.
[15] TUNG L H. Method of calculating molecular-weight-distribution functions from gel-permeation chromatograms[J]. J App Polym Sci, 1966(10):375-385.
[16] KIRKPATRICK S, GELATT C D, VECCHI M P. Optimization by simulated annealing[J]. Science,1983(220):671-680.
[17] LI C, LIU G Q, LIU Q Q, et al. Haracterization of the time evolution of starch structure from rice callus[J] Arbohydrate Polymers, 2015(127):116-123.