Redox-Inert Cations Enhancing Water Oxidation Activity: The Crucial

Aug 30, 2016 - HO and HOO ligands have consistently smaller spin densities ..... Stocker , B. D.; Sutton , A. J.; Takahashi , T.; Tilbrook , B.; van d...
0 downloads 0 Views 27MB Size
Subscriber access provided by CORNELL UNIVERSITY LIBRARY

Article

Redox-Inert Cations Enhancing Water Oxidation Activity: The Crucial Role of Flexibility Florian H. Hodel, and Sandra Luber ACS Catal., Just Accepted Manuscript • DOI: 10.1021/acscatal.6b01218 • Publication Date (Web): 30 Aug 2016 Downloaded from http://pubs.acs.org on September 2, 2016

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

ACS Catalysis is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 23

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Catalysis

Redox-Inert Cations Enhancing Water Oxidation Activity: The Crucial Role of Flexibility Florian  H.  Hodel  and  Sandra  Luber*     Department  of  Chemistry,  University  of  Zurich,  Winterthurerstrasse  190,  CH-­‐8057  Zurich,  Switzerland   Abstract:   The   devastating   effects   of   global   climate   change,   which   is   in   part   caused   by   anthropogenic   CO2   emissions   from   fossil   fuels,   force   us   to   find   clean   fuels   produced   by   environmentally   friendly   methods.   Splitting   water   into   oxygen   and   hydrogen  using  solar  light  is  one  possible  solution,  the  successful  implementation  of  which  depends  not  least  on  the  de-­‐ velopment  of  efficient  water  oxidation  catalysts  (WOCs).  With  the  water  splitting  reaction  of  photosystem  II,  specifically   the  oxygen  evolving  complex,  which  features  a  cubane  structure  with  a  redox-­‐inert  metal  center,  nature  provides  us  with   clues  for  the  construction  of  such  WOCs.  Any  approach  more  sophisticated  than  a  simple  trial-­‐and-­‐error  method  will  rely   on  knowledge  of  mechanistic  details  of  biomimetic  catalysts.  Recently,  a  step  in  that  direction  has  been  made  with  com-­‐ putational   investigations   of   the   different   possible   catalytic   pathways   of   a   {Co(II)4O4}   cubane-­‐based   WOC.   The   present   study,  which  focuses  on  the  {Co(II)3LnO4}  (Ln=Er,  Tm)  cubane  family,  is  complementary  to  the  previous  one  and  sheds   light  on  the  importance  of  redox  inert  Ln3+  cations  for  the  mechanism  of  water  oxidation.  Using  density  functional  theory   and  an  explicit  solvation  shell,  as  well  as  a  solvent  continuum  model,  the  WOCs  are  compared  in  terms  of  relative  free   energies  of  their  catalytic  states,  as  well  as  the  reaction  barriers  of  water  attack  and  oxygen  release.  Furthermore,  in-­‐depth   investigations  into  the  electronic  and  molecular  structures  of  the  catalysts  are  carried  out  resulting  in  the  discovery  of  a   flexibility  of  the  cubane-­‐cage  during  the  catalytic  cycle.     KEYWORDS:     homogeneous   water   oxidation   •   artificial   photosynthesis   •   cubane   •   density   functional   theory   •   minimum   energy  path    

Introduction Fossil  fuel  combustion  and  cement  production  were  respon-­‐ sible   for   91%   of   all   anthropogenic   CO2   emissions   between   1 2004   and   2013.   The   development   of   clean,   sustainable   fuels   and   fuel   production   methods   should   therefore   be   not   only   one   of   the   most   important,   but   also   one   of   the   most   urgent   scientific  goals  of  our  time.  Hydrogen  is  such  a  zero  emission   fuel,   either   by   itself,   or   as   an   intermediate   for   liquid   hydro-­‐ 2 carbon   fuels,   but   90%   is   produced   via   steam   reforming   of   3 natural  gas  or  light  oil.  An  alternative  to  this  environmental-­‐ ly   less   than   desirable   production   method   is   photocatalytic   water   splitting,   converting   2   water   into   2   hydrogen   and   one   oxygen  molecules.  The  efficiency  bottleneck  of  this  process  is   4 the   oxidation   half   reaction   requiring   the   transfer   of   4   elec-­‐ trons   and   protons   and   the   formation   of   an   O-­‐O   bond,   with   an  experimentally  determined  change  in  free  energy  ΔGexp:       +

-­‐

2H2O  →  O2  +  4H  +  4e  ,                                (1)     ΔGexp(pH=0,  NHE)=113.46  kcal/mol.     Nature   holds   an   abundance   of   interesting   concepts   and   mechanisms   in   store,   many   of   which   have   led   to   important   inventions  and  novel  materials  such  as  Gecko-­‐inspired  adhe-­‐ 5 6 7 sives ,  Velcro® ,  or  superhydrophobic,  self-­‐cleaning  surfaces.   It   might   therefore   be   highly   rewarding   to   apply   the   same   approach   to   the   catalysis   of   water   splitting.   Artificial   photo-­‐ 8,9 synthesis,   or,   more   precisely,   mimicking   nature’s   oxygen  

evolving  complex  (OEC)  of  photosystem  II  taking  part  in  the   light-­‐dependent   reaction   of   photosynthesis   is   a   promising   starting   point   to   develop   efficient   water   oxidation   catalysts   (WOCs).     The   OEC   is   a   CaMn4O5   cluster,   consisting   of   a   CaMn3O4   cubane   with   coordinating   amino   acids   and   a   dangling   man-­‐ 10-­‐12 ganese   connected   to   it   via   an   oxygen   atom.   It   oxidizes   water  by  going  through  5  catalytic  states  (S0-­‐S4)  constituting   13-­‐15 the   Kok   cycle.   Both   the   precise   structure   of   the   OEC,   as   well  as  the  mechanism  of  O-­‐O  bond  formation  have  however   still  not  been  completely  elucidated  and  are  subject  of  ongo-­‐ 16-­‐22 ing   research.   The   ideal   WOC   to   mimic   the   OEC   would   be   built  from  abundant,  cheap  elements  and  exhibit  not  only  a   low   overpotential,   but   also   high   turnover   numbers   (TON)   and  frequencies  (TOF).   In   2013   the   first   Co(II)-­‐cubane   based   WOC   was   synthesized   23 and   characterized   by   Evangelisti   et   al.   The   homogeneous   II catalyst   [Co 4(hmp)4(μ-­‐OAc)2(μ2-­‐OAc)2(H2O)2]   (hmp=2-­‐ (hydroxymethyl)pyridine)  (1)  (see  Figure  1)  closely  resembles   the   OEC   in   its   cubane   structure   and   ligand   environment.   Recently,  we  shed  light  on  the  mechanistic  details  of  its  cata-­‐ lytic   activity   and   suggested   possible   design   options   by   com-­‐ putationally   comparing   different   pathways   in   terms   of   ther-­‐ modynamic  free  energy  differences  as  well  as  barrier  heights   24     and  structural  properties.   In  2015  Evangelisti  et  al.  succeeded  in  improving  the  perfor-­‐ mance  of  their  catalyst  by  slightly  modifying  the  ligand  envi-­‐ ronment   and,   more   importantly,   substituting   one   Co(II)   25 center   with   a   lanthanide   cation.   The   cubane   series   II [Co 3Ln(hmp)4(OAc)5H2O];   (Ln   =   Ho   –   Yb)   have   not   only  

ACS Paragon Plus Environment

ACS Catalysis

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

been   the   first   3d-­‐4f   complexes   to   act   as   photochemical   3+ WOCs,  but  also,  by  virtue  of  the  redox  inert  Ln  cation  mim-­‐ 2+ icking  Ca ,  have  shown  even  closer  OEC  analogies  than  1.     Kohn-­‐Sham  density  functional  theory  (DFT)  based  molecular   dynamics   (MD)   simulations   have   revealed   that   the   acetate   ligands   of   the   {Co3ErO4}-­‐cubane   are   thermodynamically   far   25 less  stable  than  the  ones  of  1.  They  are  preferably  replaced   with  hydroxide,  or,  in  the  case  of  ligands  bridging  two  metal   centers,   detached   from   one   of   them,   and   the   newly   created   empty  coordination  site  is  filled  by  water.    

Page 2 of 23

II

[Co 3Ln(hmp)4(OAc)5H2O]   (Ln=Er,   Tm)   and   replaced   all   non-­‐bridging   acetate   ligands   of   the   original   molecule   with   hydroxide,  which  provided  us  with  the  catalytic  ground  state   of   our   catalysts,   hereafter   referred   to   as   2-­‐Er   and   2-­‐Tm,   re-­‐ spectively   (see   Figure   1   and   SI,   Ground   State   Structure   for   a   more  detailed  discussion).     While   2-­‐Tm   was   reported   to   show   the   lowest   catalytic   activi-­‐ ty   of   the   series   in   terms   of   TONs   and   TOFs,  2-­‐Er   resided   at   25 the   upper   end   of   the   spectrum.   Our   findings   complement   24   our   previous   study   on   the   catalytic   mechanisms   of   1, and     together   they   constitute another   essential   step   forward   to-­‐ wards  the  mechanistic  understanding  of  biomimetic  WOCs.  

Methods

Figure  1.  Structure  of  1  (top)  and  2-­‐Er/2-­‐Tm  (bottom)  and   labeling   of   the   cobalt   centers,   two   bridging   oxygen   at-­‐ oms,   and   the   “active”   ligands   “a”   and   “b”.   The   Co   center   oxidized  during  the  catalytic  cycle  is  Co1.  Additional  solvent   molecules  are  omitted  for  the  sake  of  clarity.     In  the  present  study,  we  set  out  to  investigate  the  mechanism   by   which   {Co3LnO4}-­‐cubanes   catalyze   water   oxidation   using   DFT  calculations  with  explicit  as  well  as  implicit  solvation  to   determine   not   only   free   energy   differences   between   the   states  (S0-­‐S4)  of  the  catalytic  cycle  (which  will  be  described   in  section  Results,  Mechanism,  specifically  in  Scheme  1),  but   also   minimum   energy   paths   and   barriers.   To   this   end,   we   focused   our   attention   on   two   catalysts   of   the   series  

The   methods   we   used   are   to   a   large   extent   identical   to   the   ones   we   had   employed   and   described   in   our   study   of   the   catalytic   pathways   of   1,   where   we   had   also   justified   our   24 choices   of   procedures,   set-­‐ups,   and   parameters.   We   there-­‐ fore   give   here   only   a   brief   overview,   especially   highlighting   methodological  differences  from  the  earlier  study.     We   used   two   different   and   somewhat   complementary   ap-­‐ proaches   to   obtain   structures,   electronic   energy   differences   and   other   properties   of   the   catalytic   states.   Employing   the   26 CP2K  program  package,  we  performed  geometry  optimiza-­‐ tions  and  single  point  energy  calculations  of  all  states  includ-­‐ ing  an  explicit  water  shell  of  68  molecules.  Because  the  struc-­‐ tural   differences   between   the   states   are   small   and   geometry   optimizations   starting   from   different   water   shell   structures   led  to  similar  results,  we  reasoned  that  this  method  is  accu-­‐ rate   enough   for   our   purposes   and   molecular   dynamics   or   24 other   sampling   techniques   were   not   necessary.   Neverthe-­‐ less,   neglecting   such   atomistic-­‐level   fluctuations   of   the   sol-­‐ vent  constitutes  an  approximation  made  in  this  approach.     27 Using  the  Turbomole  6.5  program   package,  we  carried  out   calculations  of  the  same  systems,  this  time  however  with  the   28 conductor-­‐like  screening  model  (COSMO).  While  an  obvi-­‐ ous   shortcoming   of   this   approach   is   the   neglect   of   short-­‐ range  solute-­‐solvent  interactions,  such  as  hydrogen  bonding,   it   is   computationally   cheaper   and   provides   “averaged”   sol-­‐ 29 vent  effects.     The   calculations   carried   out   with   CP2K   and   explicit   solvent   consisted   of   geometry   optimizations   with   the   BP86   ex-­‐ 30,31 change-­‐correlation   density   functional,   followed   by   single   point   energy   calculations   with   the   B3LYP   hybrid   density   32,34 functional.  The  initial  configurations  of  the  S0  states  were   extracted   from   a   DFT-­‐based   MD   run   of   2-­‐Er   by   deleting   excess   water   molecules   (see   SI,   Ground   State   Structures   for   details).  The  initial  configurations  of  the  geometry  optimiza-­‐ tions  of  the  other  states  were  obtained  by  removing  a  proton   and  an  electron  from  the  previous  state  (and  forming  an  O-­‐O   bond   with   a   solvent   shell   water   molecule   for   S3).   For   the   systems   with   implicit   solvation,   we   additionally   performed  

ACS Paragon Plus Environment

Page 3 of 23

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Catalysis

geometry   optimizations   with   B3LYP.   In   all   calculations,   we   35 employed  Grimme’s  D3  dispersion  correction.   The   set-­‐up   of   the   CP2K   calculations,   which   used   the   36 QUICKSTEP   program,   consisted   of   mixed   Gaussian   and   plane   wave   basis   sets   in   combination   with   (relativistic)   25,37,38 Goedecker-­‐Teter-­‐Hutter   (GTH)   pseudopotentials,   39 DZVP-­‐MOLOPT-­‐GTH  basis  sets  (DZVP-­‐MOLOPT-­‐SR-­‐GTH   for   Co,   and   a   double-­‐zeta   valence   basis   set   for   Er   and   25,38 Tm ),  and  a  plane  wave  cutoff  of  400  Ry.  The  systems  were   3 contained  in  a  30  Å3  box.   40,41 The  Turbomole  calculations  used  def2-­‐TZVP  basis  sets,  a   scalar  relativistic  effective  core  potential  (ECP-­‐28)  for  Er  and   Tm,   and   the   resolution-­‐of-­‐the-­‐identity   density-­‐fitting   tech-­‐ 42   43 nique with  corresponding  auxiliary  basis  sets.     In   all   catalytic   states,   the   Co   and   Er/Tm   centers   were   as-­‐ sumed   to   couple   ferromagnetically   with   each   other   and,   except   for   Co1   (for   the   numbering   of   the   metal   centers   see   24,25,44 Figure   1),   to   be   in   a   high-­‐spin   state.   While   for   S0,   we   took   Co1   to   be   in   a   quartet   state,   for   S1-­‐S4,   we   determined   the   local   minimum   energy   configuration   and   the   associated   electronic  energy  for  three  different  total  multiplicities  corre-­‐ sponding  to  three  possible  spin  states  of  Co1.  The  actual  spin   configurations   on   the   metal   centers   and   the   “active”   oxygen   ligands   were   monitored   with   Mulliken   spin   populations,   which  however  depend  on  the  basis  set  and  in  general  do  not   45 converge   to   a   basis   set   limit.   Nevertheless,   Mulliken   spin   population   analysis   is   routinely   applied.   Following   the   ap-­‐ 46-­‐49 proach  by  Nørskov  et  al.,  we  then  calculated  free  energy   differences   ∆G!!!   between   a   state   i   and   the   catalytic   ground   state  as    

∆G!!! = E! + E!"#,! + 0.5iE!! + 0.5iE!"#,!! − E! − E!"#,! + i ∆H − T∆S  ,      i = 1, … ,4.          (2)     Ei  denotes  the  electronic  energy  of  the  state  Si,  EH2  the  elec-­‐ tronic   energy   of   a   hydrogen   molecule,   E0   the   electronic   ener-­‐ gy   of   S0   and   EZPE   the   corresponding   zero   point   energies,   which  we  obtained  from  normal  mode  analyses  (see  SI,  Table   S3)   employing   the   setup   described   above.   ΔS=0.016   kcal/(K*mol)  is  half  the  entropy  of  H2  at  standard  conditions   50 taken   from   thermodynamic   tables.   Similarly,   the   enthalpy   can  be  calculated  as  ΔH=1.153  kcal/mol.  For  the  systems  with   implicit   solvent,   where   the   attacking   water   molecule   does   not   originate   from   the   solvation   shell,   the   free   energy   of   a   single   water   molecule   needs   to   be   included   in   equation   (2)   24 for  states  S3  and  S4.     We   approximated   the   overpotential   η   using   the   largest   free   energy   difference   between   two   consecutive   states   of   the   !"# catalytic  cycle   ΔG!!(!!!)   and  the  computational  value  of  the   reaction  free  energy  of  equation  (1),   ΔG!"#$ (H2O),  (see  sec-­‐ 46,51 tion  Free  Energy  Differences)  according  to      

!"# η=ΔG!!(!!!) −

Δ!!"#$ (H2O) 4

             

 

       

 (3)  

For   the   calculation   of   minimum   electronic   energy   reaction   paths   and   the   barriers   associated   with   them,   we   used   the   nudged  elastic  band  (NEB)  method  as  implemented  in  CP2K.   The  calculations  of  electronic  energies  and  forces  were  done   exactly   the   same   way   as   for   the   geometry   optimizations   de-­‐ scribed   above.   Due   to   high   computational   cost   of   hybrid   functionals,   only   the   BP86   functional   was   employed,   which,   however,  tends  to  describes  electron  distribution  in  too  delo-­‐ 52,53 calized   way   compared   to   B3LYP.   Except   where   noted   otherwise,   all   NEB   calculations   consisted   of   8   replicas   or   54   frames  and  were  optimized  by  a  climbing  image  NEB every   th 55 5   step   and   the   improved   tangent   method   for   all   others.   The  initial  guesses  for  the  intermediate  frames  were  obtained   by   linear   interpolation   of   the   atom   positions.   We   assumed   every   replica   to   have   the   same   total   number   of   unpaired   electrons  as  the  first  one.  For  those  first  frames,  we  chose  the   BP86   minimum   electronic   energy   spin   multiplicities,   which   can   differ   from   the   B3LYP   minimum   electronic   energy   multi-­‐ plicities   used   for   the   calculations   of   free   energy   differences   (see  SI,  Tables  S1  and  S2).  To  compare  the  reaction  paths  of   the   two   cubanes   only   considering   differences   stemming   directly   from   the   (electronic)   structure   of   the   catalysts   them-­‐ selves,   disentangled   from   contributions   of   solvent   interac-­‐ tions,   we   performed   the   NEB   calculations   of   the   O2   release   from  2-­‐Er  and  2-­‐Tm  in  vacuo.   For   the   visualization   of   the   molecular   structures   we   used   56 VMD  1.9.1.  

Results   Mechanism.   Our   investigations   of   the   catalytic   cycle   of   1   had   focused   on   two  catalytic  pathways,  a  single-­‐site  and  an  oxo-­‐oxo  coupling   24 mechanism   involving   two   Co   centers.   Assuming   that   all   ligands   in   the   structure   of   2   (see   Figure   1)   are   stable,   it   is   obvious  that  a  pathway  with  two  participating  metal  centers   is  impossible  for  this  catalyst  due  to  the  absence  of  water  or   hydroxide  ligands  or  free  coordination  sites  on  Co2  and  Co3.   The  only  possible  coupling  partner  for  an  oxo  ligand  on  Co1   is  a  hydroxide  ligand  attached  to  the  same  Co1  or  the  lantha-­‐ nide   atom.   While   geminal   coupling   of   two   non-­‐bridging   57 ligands   on   the   active   Co   would   warrant   further   research   3+ beyond  this  study,  Ln  is  thought  to  behave  as  a  redox-­‐inert   2+ 25 analog   to   Ca   in   the   OEC.   We   therefore   did   not   consider   any   reactions   requiring   the   direct   participation   of   Er   or   Tm   and   refer   the   reader   to   our   study   of   the   mechanism   of   1   where  we  have  discussed  (and  ruled  out)  a  number  of  other   24 possible   mechanisms.   Thus,   we   are   left   with   a   single-­‐site   pathway  (see  Scheme  1)  with  Co1  as  the  active  metal  center.   We   assume   4   consecutive   proton-­‐coupled   electron   transfer   (PCET)   steps   between   the   5   states   (S0-­‐S4)   of   the   catalyst   and   that  the  O-­‐O  bond  formation  takes  place  as  a  water  molecule   from   the   solvent   shell   attacks   the   oxyl   ligand   of   S2   prior   to  

ACS Paragon Plus Environment

ACS Catalysis 58

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

the  PCET.  Due  to  the  lower  acidity  of  the   hydroxide  ligand,   the  initial  step  of  the  single-­‐site  mechanism  is  the  deprotona-­‐ tion   of   the   water   ligand   and   the   concurrent   oxidation   of   Co(II)   to   Co(III)   resulting   in   a   dihydroxo   species   (S1).   The   proton  of  the  second  PCET  could  originate  from  either  OH-­‐ ligand,  resulting  in  the  bifurcation  of  the  pathway  (denoted  a   and  b  in  Scheme  1).  

Page 4 of 23

the  thermodynamic  limit  of  water  oxidation  (upper  graph  of   Figure   2   and   SI,   Figures   S1   and   S2,   Table   S4).   Secondly,   based   60 on  the  Sabatier  principle  and  the  idea  that  a  thermodynam-­‐ ically   ideal   catalyst   would   exhibit   equal   free   energy   differ-­‐ ences   between   its   states,   we   compared   the   various   systems   with   such   an   ideal   catalyst   (lower   graph   of   Figure   2   and   SI,   61,62 ope  S2).  Finally,  we  calculated  overpotentials  by  approxi-­‐ mating  them  using  the  largest  free  energy  difference  between   two   consecutive   states   of   the   catalytic   cycle   (SI,   Table   S5),   an   46,51 approach  introduced  by  Nørskov  et  al.  

Scheme  1.  Catalytic  cycle  of  2-­‐Er  and    2-­‐Tm.  Only  Co1  and   the  ligands  actively  participating  in  the  reactions  are  shown.     For   both   S2   states,   two   resonance   structures   exist   (only   drawn   for   S2a   in   Scheme   1):   Co(IV)   with   an   oxo   ligand   and   Co(III)  and  an  oxyl  radical,  which  has  been  shown  for  other   59 systems   to   facilitate   O-­‐O   bond   formation   and   had   also   24   computationally   been   found   to   be   favored   by   1. Next,   a   nucleophilic   attack   by   a   water   molecule   takes   place   and   is   followed   by   the   third   PCET   to   form   S3a   or   S3b.   Finally,   the   fourth  proton  and  electron  are  lost  leading  to  the  S4a  or  S4b   state.   As   for   S2,   resonance   structures   with   radical   character   on  the  ligand  can  be  envisioned.  We  had  shown  that  1  prefers   a   configuration   which   is   best   described   by   resonance   struc-­‐ 24 tures  similar  to  the  middle  and  right  one  in  Scheme  1.  The   same   holds   true   for   2-­‐Er   and   2-­‐Tm.   The   final   step   restoring   the   catalyst   to   its   ground   state   is   the   release   of   O2   and   the   uptake   of   a   water   molecule.   The   structure   of   each   catalytic   state   can   be   found   in   the   SI   (Figures   S4-­‐S6).   Overall,   as   re-­‐ + quired   by   equation   (1),   O2,   4H   and   4   electrons   have   been   produced  from  2  water  molecules.   Free  Energy  Differences.   We   calculated   free  energy  differences  according  to  equation   (2)   (for   now   considering   only   the   spin   multiplicities   of   each   state   with   the   lowest   electronic   energy)   and   compared   the   two   different   lanthanide-­‐containing   cubanes   and   the   two   24 cubane-­‐types   (2   and   1 )   (for   results   obtained   with   varying   computational  methods,  see  SI,  Relative  (Free)  Energies).  To   this   end,   we   processed   and   analyzed   the   free   energies   in   three  ways:  Firstly,  we  plotted  free  energy  profiles  including  

Figure   2.   Free   energy   differences   between   the   states   i   and   state   0   of   the   catalytic   cycle.   (B3LYP;   explicit   solva-­‐ tion;  for  results  obtained  with  other  set-­‐ups,  see  SI,  Figures  S1   and  S2  and  Table  S4).  The  dotted  lines  and  bars  correspond   to  the  “b-­‐states”.  Upper  graph:  Relative  free  energies  and  the   computed  thermodynamic  limit.  Lower  graph:  Free  energies  

ACS Paragon Plus Environment

Page 5 of 23

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Catalysis

compared   to   an   “ideal”   catalyst   [scaled   by   ΔGexp(H2O)   /   ΔGcomp(H2O)].       We   plotted   ΔGi-­‐0,   the   free   energy   difference   between   state   i   and  the  catalytic  ground  state,  calculated  with  CP2K,  B3LYP   and   explicit   solvation,   in   the   upper   graph   of   Figure   2   (for   results   obtained   using   COSMO,   see   SI,   Figure   S2)   The   ther-­‐ modynamic  limit  in  this  graph  is  not  the  experimental  reac-­‐ tion  free  energy  ΔGexp(H2O)  of  equation  (1),  but  the  computa-­‐ tional   one   (ΔGcomp(H2O)=102.6   kcal/mol)   obtained   with   the   61 same   set-­‐up   as   the   states   of   the   catalytic   cycle.   Graphs   comparing   free   energy   profiles   obtained   with   implicit   solva-­‐ tion  and  other  exchange-­‐correlation  functionals  can  be  found   in  Figure  S1  in  the  SI.  From  the  upper  graph  of  Figure  2,  we   can  see  that  pathway  b  lies  lower  than  pathway   a,   and  2-­‐Er  is   below  2-­‐Tm  with  respect  to  free  energy  differences.  The  only   exception   is   S2   where   2-­‐Tm-­‐S2b   is   slightly   below   2-­‐Er-­‐S2a,   and   2-­‐Er-­‐S2b   lies   higher   than   any   other   system   in   the   S2   state.   This   stands   in   connection   with   a   structural   change   of   the  catalysts,  best  described  as  “opening”  of  the  cubane  cage,   which   will   be   further   discussed   in   section   Electronic   Struc-­‐ ture  and  Molecular  Geometry.  Except  for  cycle   b  of  2-­‐Er,   all   pathways  exceed  the  thermodynamic  limit  already  with  their   S3  state.  While  the  single-­‐site  pathway  of  1  is  thermodynami-­‐ cally   most   favorable   for   S1   and   S2   (together   with   2-­‐Tm-­‐ 24 S2b),  for  S3  and  S4,  2-­‐Er  shows  the  lowest  free  energy  dif-­‐ ferences,  irrespective  of  the  pathway.     The  lower  graph  of  Figure  2  was  obtained  by  multiplying  the   relative   free   energies   by   1.1,   the   ratio   of   ΔGexp(H2O)   to   ΔGcomp(H2O),  to  provide  the  experimental  reaction  free  ener-­‐ gy  of  water  oxidation  and  account  for  errors  inherent  to  our   61 approach.   While   in   their   S1   state,   all   systems   are   close   to   the  ideal  catalyst,  they  diverge  more  and  more  for  S2  and  S3   and   appear   too   destabilized.   For   2-­‐Tm,   it   would   be   most   beneficial   to   stabilize   S3,   whereas   the   catalytic   cycle   of   2-­‐Er   could  be  brought  closer  to  the  ideal  case  by  lowering  S2  and   S3   in   free   energy.   As   mentioned   above,  2-­‐Er   is   the   only   sys-­‐ tem  for  which  the  last  PCET  step  and  the  evolution  of  oxygen   is  endothermic  (only  in  cycle  b).  With  implicit  solvation  (SI,   Figure  S2)  all  systems  reproduce  the  ideal  catalyst  better  than   with   explicit   solvation.   Furthermore,   the   solvent   continuum   calculations   predict   cycle   b   of   2-­‐Tm   to   come   closest   to   the   ideal  case.   Cycle   a   of   2-­‐Er   has   an   overpotential   (vs.   NHE,   pH=0)   of   η=1.0  V   associated,   whereas   the   one   of   2-­‐Tm   amounts   to   η=0.8  V.  For  both  catalysts,  the  highest  free  energy  difference   of  cycle  a  is  between  S2a  and  S3a.  The  overpotentials  of  cycle   b  of  the  two  cubanes  is  higher,  η=1.2  V  (S1→S2b)  for  2-­‐Er  and   η=1.3  V   (S2b→S3b)   for   2-­‐Tm.   Calculations   with   implicit   solvation  consistently  give  lower  free  energy  differences  (see   SI,  Figure  S1  and  Table  S4)  and  therefore  also  lower  overpo-­‐ tentials  (SI,  Table  S5).  A  more  detailed  analysis  of  the  influ-­‐ ence   of   different   functionals   and   solvation   methods   on   the  

relative   free   energies   can   be   found   in   the   SI,   Relative   (Free)   Energies.     Total  Spin  Multiplicities.   In  order  to  investigate  the  impact  of  different  spin  configura-­‐ tions   on   the   “active”   Co1   center,   we   preformed   all   geometry   optimizations   of   S1-­‐S4   separately   for   the   3   total   spin   multi-­‐ plicities   corresponding   to   an   initial   guess   of   high-­‐spin   on   Co2,   Co3   and   Ln   and   3   different   numbers   of   unpaired   elec-­‐ trons   on   Co1.   In   S1   and   S3   of   2-­‐Er,   these   numbers   were   0,2   and  4  amounting  to  total  spin  multiplicities  of  M=10,12,14.  In   S2  and  S4  of  2-­‐Er,  the  number  of  unpaired  electrons  on  Co1   was   1,3,   and   5   corresponding   to   total   spin   multiplicities   of   M=11,13,15.   For   2-­‐Tm   all   total   multiplicities   are   lower   by   1   since   Tm   possesses   one   electron   more   than   Er.   It   has   to   be   kept   in   mind   that   predictions   of   low-­‐spin-­‐high-­‐spin   energy   splittings   are   not   a   strong   point   of   DFT.   In   particular,   pure   DFT   functionals   tend   to   predict   low-­‐spin   states   to   have   a   lower   electronic   energy   than   higher   spin   states,   whereas   hybrid   functionals,   such   as   B3LYP,   usually   favor   high-­‐spin   63 states.  This  depends  however  strongly  on  the  percentage  of   64 exact  exchange  admixed.   It   can   be   seen   from   Tables   S1-­‐S2   and   S22-­‐S23   in   the   SI   that,   with   explicit   solvent   and   B3LYP,   S1   favors   the   intermediate   spin   multiplicity,   whereas   the   lowest   electronic   energy   mul-­‐ tiplicities   of   S2-­‐S4   (of   2-­‐Er   and   2-­‐Tm)   are   high-­‐spin   (except   for  S2a  of  2-­‐Er  which  is  predicted  to  prefer  the  intermediate   spin  multiplicity).  BP86  on  the  other  hand  predicts  only  S2b   of   both   catalysts   to   clearly   prefer   a   high-­‐spin   configuration   (S4a  of  2-­‐Er  also  favors  high-­‐spin,  but  the  low-­‐spin  configu-­‐ ration  is  only  2.5  kcal/mol  higher  in  electronic  energy).  These   results  point  in  the  direction  of  the  trend  mentioned  above.   However,   in   our   study   of   1,   we   had   found   (using   the   exact   same  methods)  that  low  spin  multiplicities  were  predicted  by   B3LYP   for   all   states   of   a   catalytic   cycle   (except   for   one,   which   differed  by  only  a  few  kcal/mol  from  the  low-­‐spin  configura-­‐ 24   tion). Furthermore,  in  that  study,  we  had  repeated  all  single   point  calculations  with  B3LYP*  and  had  found  no  difference   in   the   energetic   ordering   of   the   spin   states   and   virtually   no   change   in   electronic   energy   differences.   Finally,   calculations   with  implicit  solvation  predicted  only  a  few  states  to  be  high-­‐ spin,   irrespective   of   the   functional   employed   (see   SI,   Tables   S1-­‐S2  and  S22-­‐S23).     While   the   lowest   electronic   energy   multiplicities   depend   on   the   computational   set-­‐up,   they   differ   only   slightly   between   the   two   catalysts   investigated.   With   explicit   solvent   and   B3LYP,  only  S2a  shows  a  difference  (intermediate  multiplici-­‐ ty  for  2-­‐Er  and  high-­‐spin  for  2-­‐Tm).  With  the  other  methods,   the   picture   is   similar   and   most   states   show   the   same   lowest   electronic  energy  multiplicity  for  2-­‐Er  and  2-­‐Tm.     According   to   calculations   using   explicit   solvent   and   B3LYP,   both   cycles   of   both   catalysts   contain   2   spin   crossing   events.   However,   it   should   be   kept   in   mind   that   no   spin-­‐orbit   cou-­‐

ACS Paragon Plus Environment

ACS Catalysis

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 6 of 23

pling   is   included   in   our   calculations,   which   might   facilitate   spin-­‐state   crossing,   in   particular   in   the   presence   of   the   lan-­‐ thanide  metal  centers.   Electronic  Structure  and  Molecular  Geometry.   Both   exchange-­‐correlation   functionals,   solvation   methods,   catalysts,   and   all   total   multiplicities   have   in   common   that   they  show  large  spin  populations  on  the  O-­‐  and  OO-­‐ligands   of   the  S2   and   S4   states,   respectively.   HO-­‐   and   HOO-­‐ligands   have   consistently   smaller   spin   densities   associated   (see   SI,   Tables   S7-­‐S20).   This   agrees   with   our   findings   on   the   Co4-­‐ 24   cubane   catalyst and   many   other   studies   on   different   sys-­‐ tems,  which  found  the  oxyl  radical  to  play  an  important  role   in   catalysis   by   enhancing   the   reactivity   of   the   respective   59,65-­‐67 intermediate.   Throughout   the   catalytic   cycle,   spin   density  was  not  only  found  on  the  metal  atoms,  but  also  on   all   oxygen   atoms   bridging   them.     Furthermore,   all   calcula-­‐ tions  agreed  that  the  spin  density  on  the  lanthanides  stayed   the  same  in  every  state,  thus  underscoring  the  role  of  Er  and   2+ Tm  as  redox  inert  Ca -­‐analogs.   Next,   we   turned   to   the   differences   in   molecular   geometries   between  the  two  catalysts  (see  SI,  Figure  S8).  Our  investiga-­‐ tions  of  the  structure  of  the  solvation  shell  and  the  position   of   the   ligands   are   described   in   great   detail   in   the   SI   (Struc-­‐ tural  Analysis).  Here  we  focus  on  another  effect:  the  confor-­‐ mation  of  the  cubane-­‐cage  itself.  As  already  hinted  at  in  our   discussion   of   free   energy   differences,   for   some   states   the   cubane   is   “opened”,   i.e.   the   distances   between   Co1   and   the   two   oxygen   atoms   (O1   and   O2)   that   bridge   Co1   to   Co2   (see   Figure   1)   are   increased.   While   for   2-­‐Er,   the   Co1-­‐O1   and   Co1-­‐ O2   distances   are   smaller   in   S1   and   S2a   than   in   S0,   they   are   larger  for  2-­‐Tm  (see  Table  1).  In  S2b,  the  cubane  cage  of  both   2-­‐Er  and  2-­‐Tm  is  significantly  “opened”;  the  effect  is  however   more  pronounced  for  2-­‐Tm.  After  the  water  attack,  there  are   some   fluctuations   in   the   molecular   geometry   of   the   catalysts.   However,  when  looking  at  the  sum  of  the  Co1-­‐O1  and  Co1-­‐O2   distances   in   S3   and   S4,   it   becomes   apparent   that   firstly,   the   molecular   geometry   is   similar   for   2-­‐Er   and   2-­‐Tm,   and   sec-­‐ ondly,   it   is   larger   for   the   “a-­‐states”   than   for   the   “b-­‐states”   (Figure   3).   Hence,   there   is   no   significant   difference   in   the   sum   of   the   two   bond   lengths,   describing   the   “opening”   of   the   cubane  cage,  between  2-­‐Er  and  2-­‐Tm  in  states  S0  and  S3-­‐S4.   The   pronounced   differences   observed   for   states   S1,   S2a   and   S2b,   on   the   other   hand,   go   along   with   a   difference   in   frontier   orbitals  between  the  two  cubanes  (see  below).  

Figure   3.   Sums   of   the   Co1-­‐O1   and   Co1-­‐O2   distances   for   both  catalysts.    

Table   1.     Distances   (in   Å)   between   Co1   and   the   two   bridging  oxygen  atoms  O1  and  O2  .    

2-­‐Er  

2-­‐Tm  

Co1-­‐O1  

Co1-­‐O2  

Co1-­‐O1  

Co1-­‐O2  

S0  

2.20  

2.10  

2.18  

2.10  

S1  

2.01  

1.99  

2.25  

2.20  

S2a  

2.06  

2.05  

2.22  

2.41  

2.49  

2.22  

2.67  

2.21  

2.1-­‐2.3  

2.1-­‐2.2  

2.1-­‐2.3  

2.1-­‐2.2  

S2b   a

S3-­‐S4   a  

The  sum  of  the  two  bond  lengths  in  S3a,  S3b  ,S4a,  and  S4b   are  similar  for  2–Er  and  2-­‐Tm.   Molecular  geometries  obtained  with  Turbomole  and  COSMO   are  in  most  cases  not    significantly  influenced  by  the  choice   of  the  functional  (BP86  or  B3LYP)  employed  for  the  geome-­‐ try   optimizations   (see   SI,   Relative   (Free)   Energies   for   more   details).   The  main  difference  in  molecular  geometries  between  calcu-­‐ lations  with  CP2K  including  explicit  solvation  and  Turbomo-­‐ le   in   combination   with   COSMO   is   that   with   the   former,   ligands   often   interact   and   hydrogen   bond   with   water   mole-­‐ cules   of   the   solvation   shell   while   with   the   latter,   they   often   interact  with  other  ligands.  For  all  states,  this  influences  the   positions   and   orientations   of   the   bridging   acetates   and   the   OH   ligands   on   Ln,   which   appear   to   be   less   tightly   bound   and   therefore   more   susceptible   to   solvation   effects   than   the   hy-­‐ droxide   ligands   on   Co1   (bond   lengths   of   Ln-­‐OH:   2.1-­‐2.2   Å,   Co-­‐OH:   1.8-­‐1.9   Å   for   both   explicit   and   implicit   solvation).   A  

ACS Paragon Plus Environment

Page 7 of 23

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACS Catalysis

more  detailed  analysis  of  the  molecular  geometries  resulting   from  different  treatments  of  the  solvent  can  be  found  in  the   SI   (Structural   Analysis).   From   these   investigations,   it   is   ap-­‐ parent   that   short   range   effects,   hydrogen   bonding   and   the   inclusion   of   explicit   solvation   is   most   important   for   the   fol-­‐ lowing   cases:   Weakly   bound   ligands   (Ln-­‐OH),   ligands   being   close   enough   to   other   ligands   to   preferentially   interact   with   them   instead   of   the   continuum   (Co1-­‐OH   and   Ln-­‐OH   in   the   “a-­‐states”)   and   larger   ligands   with   more   conformational   degrees   of   freedom   and   more   potential   sites   for   solute-­‐ solvent  interactions  (Co1-­‐OOH)  protruding  far  into  the  con-­‐ tinuum  or  the  solvation  shell  (bridging  acetate).   It  can  be  concluded  that  firstly,  the  “opening”  of  the  cubane   is   due   to   short-­‐range   solvation   effects   since   the   calculations   with   implicit   solvation   show   no   such   distortions   and   only   very   small   (