**1. Introduction**

Furfural is an important biomass-based high-value chemical with numerous applications [1]. While it is widely used directly as a solvent and a fungicide, it is most commonly converted into pharmaceuticals, chemicals and biopolymers, many of which are used as substitutes for petrochemical-derived analogs [2,3].

Approximately 300,000 tons of furfural are produced annually and the biggest manufacturer is China [4]. It is formed by the dehydration of pentoses, mainly xylose, which can be obtained through the hydrolysis of different agricultural residues (corn stover, wheat straw, sugarcane bagasse, rice husk, oat hull) or forest industry wastes (birch or poplar sawdust). Current industrial processes use mineral acids, such as sulfuric, phosphoric or hydrochloric acid, as catalysts, with an approximate furfural yield of 50% [5,6]. The relatively low yield, high energy consumption (caused by the high, 150–240 ◦C, reaction temperature) and environmental concerns related to acidic process wastes are driving scientists to develop catalysts with high selectivity for furfural formation. A number of studies have used inexpensive water-soluble inorganic salts (mainly chlorides) as catalysts instead of mineral acids for xylose conversion to furfural [7–10]. Many of those studies show that FeCl3 results in the highest furfural yields compared to other metal chlorides [7,10].

Compared to water-soluble salts, solid catalytic systems have fewer environmental impacts than liquid ones, reduce operational costs and are technically feasible alternatives for industry [11]. A broad

range of different heterogeneous catalysts have been studied for xylose conversion to furfural, the most common of which are acidic zeolites and mesoporous silicas, such as SBA-15 and MCM-41 [11,12]. Additionally, sulfonated metal oxides (especially TiO2, ZrO2) have been used, either as such or combined with mesoporous supports [13–16]. Many studies have shown that some Brønsted acidity is needed, either from metal oxide or support, in addition to metal oxides with a Lewis character, to achieve good furfural yields [17,18]. The need for Brønsted acidity is associated with the furfural production mechanism; while Lewis acid can isomerize xylose to xylulose, Brønsted acid is needed for the dehydration step (Scheme 1) [19].

**Scheme 1.** Dehydration of xylose to furfural by Brønsted acid and Lewis acid catalysts.

Carbon-based catalysts are attractive since the carbon supports are low-cost materials with high surface area and good thermal stability and they are easily modified with functional groups [20]. Carbon surface groups containing heteroatoms, such as oxygen, can act as anchoring sites for metal particles and generate high metal dispersion [21]. The diverse surface also enables Brønsted acidity in addition to metal's Lewis acidity, which can further promote xylose conversion to furfural. To the best of our knowledge, few papers have been published on carbon-supported metal oxide catalysts. Mazzotta et al. used a sulfonated carbonaceous material with TiO2 sites, while Russo et al. used TiO2/carbon black in the conversion of xylose to furfural [22,23]. Barroso-Bogeat et al. prepared a class of different activated carbon-metal oxide catalysts (Fe, Al, Zn, Sn, Ti and W) but did not use them in any reaction [24–26]. Therefore, cheap metal oxides, such as iron oxide, have not been utilized in xylose conversion to furfural, even though iron is common in homogeneous catalysts. The aim of this study was to create iron oxide sites on a carbon-supported catalyst and apply the catalyst in the conversion of xylose to furfural. Activated carbon support was derived from hydrolysis lignin, which is a side stream of cellulosic ethanol production. Reactions were performed in biphasic media to increase reaction selectivity.

#### **2. Results and Discussion**

## *2.1. Preliminary Studies*

The purpose of the preliminary studies was to optimize the reaction media and select the best catalytic metal for heterogeneous catalysts.

#### 2.1.1. Furfural Partitioning in Biphasic Reaction System

The reaction system was optimized in terms of the appropriate organic solvent and solvent:water ratio. Experiments were carried out using a 4.7 wt% furfural solution in water as feed and toluene or methyl isobutyl ketone (MIBK) as an organic solvent. The purpose was to compare furfural partitioning from water into these two solvents. Solvent ratios of 1:1, 1:2 and 1:3 were used based on the literature [17,27,28]. The results of partitioning experiments are shown in Figure 1, and, as expected, the more organic solvent was added, the better furfural was extracted to the organic phase (Figure 1a). MIBK showed a better ability to extract furfural than toluene, as the furfural content in water was

relatively higher in the toluene experiments than in the MIBK experiments. In the best case, when 1:3 water:MIBK was used as a solvent, 96% of furfural was extracted to the MIBK phase. Similarly, MIBK was reported to extract furfural better than toluene in the literature [27,29]. MIBK has a polar carbonyl group structure that may interact better with furfural, which is an aldehyde, compared to nonpolar toluene. The partitioning coe fficient of MIBK has been reported to be 7 [29], while in our study it was 8 using a similar solvent ratio (1:1) and 8.5 when the solvent ratio was 1:3 (Figure 1b). For toluene, the partitioning coe fficient has been reported to be 3 [29] but in our study it was significantly higher (6) with a 1:1 solvent ratio. Di fferences in reported and measured partitioning coe fficients are likely due to di fferent experimental setups. In the cited study, the partitioning coe fficients were determined without any heating, using only shaking as an extraction technique. In our study, 5 min microwave heating (at 160 ◦C) was used, as the higher temperature corresponds to the reaction conditions in conversion. Based on the partitioning results, MIBK was considered most suitable for the furfural removal from water and it was used as an organic solvent in further experiments.

**Figure 1.** Furfural partitioning between water and methyl isobutyl ketone (MIBK) and water and toluene using di fferent solvent ratios. (**a**) Partitioning calculated as the ratio of moles of furfural in the organic phase and in the water phase; (**b**) Partitioning calculated as the partitioning coe fficient (based on concentrations).

#### 2.1.2. Furfural Production Using Homogeneous Catalysts

After choosing the reaction media, homogeneous metal salts (ZnCl2, AlCl3·6H2O, CrCl3·6H2O, SnCl2·2H2O and FeCl3·6H2O) were used as catalysts to determine the best metal for furfural production. Di fferent metal salts have been studied in the literature but it was important to perform tests in our reaction system, using microwave heating and water/MIBK media. Reactions were performed at a temperature of 160 ◦C, a reaction time of 1.5 h and a catalyst amount of 0.05 mmol, based on the literature [10]. Almost full xylose conversion (98%–99%) was achieved using all catalysts other than zinc chloride, which yielded only 91% conversion (Figure 2, asterisks). In the furfural yields, there was variation between catalysts and the yields varied between 33% and 68% (Figure 2, bars). Chromium chloride and aluminum chloride produced the lowest furfural yields, although the xylose conversion was almost complete. Therefore, their selectivities (34% and 38%, respectively) were poorest among all catalysts (Figure 2, squares). Tin chloride and zinc chloride produced slightly higher furfural yields compared to chromium or aluminum and their selectivities were also higher (46% and 48%, respectively). However, the significantly highest furfural yield (68%) and selectivity (70%) were achieved using iron chloride as a catalyst. Our result agrees with the literature, as many studies have shown that FeCl3 produces the highest furfural yield in water media compared to other metal

halides [7,10]. The 68% furfural yield achieved in our study with FeCl3 catalyst is also comparable to that found in other studies. For example, Ershova et al. achieved a 60.3% furfural yield in water at 180 ◦C, while Zhang et al. reported a 77% furfural yield in gamma-valerolactone at 160 ◦C [10,30]. However, in most studies NaCl has been used as a phase modifier, and, because Cl− has been shown to catalyze furfural formation, it is difficult to compare those results to our study without NaCl [31–34].

**Figure 2.** Furfural yield (bar), respective conversion (asterisk) and reaction selectivity (square) during xylose dehydration at 160 ◦C for 1.5 h using different metal chlorides (0.05 mmol) as catalysts.

#### *2.2. Preparation and Characterization of Heterogeneous Catalysts*

After the appropriate reaction media and catalytic metal were chosen, three different catalyst supports (ACz, ACzN, ACs) and five different catalysts (5FeNO3-ACz, 5Fe-ACz, 5Fe-ACzN, 5Fe-ACs and 10Fe-Acs) were prepared (Table 1). In the preparations, different activation methods, metal precursors, metal concentrations and additional treatments were used. ACz support was prepared applying common ZnCl2 chemical activation [35]. ACz-based catalysts 5FeNO3-ACz and 5Fe-ACz were prepared using 5 wt% of either FeNO3 or FeCl3 as an iron precursor, respectively. ACzN support was prepared using HNO3 treatment after ZnCl2 chemical activation and ACzN-based catalyst 5Fe-ACzN by further impregnation with FeCl3. The third support type, ACs, was prepared using the physical activation method. Compared to chemical activation, physical activation is performed at a higher temperature but only using water steam as an activation agen<sup>t</sup> [21,36]. ACs-based catalysts 5Fe-ACs and 10Fe-ACs were prepared using FeCl3 as a metal precursor and 5 or 10 wt% as the metal concentration.

**Table 1.** Key factors in the preparation of various supports and catalysts and the metal contents (wt%) of the prepared catalysts measured by inductively coupled plasma optical emission spectrometry (ICP-OES).


The metal contents (according to inductively coupled plasma optical emission spectrometry (ICP-OES)) of all supports and catalysts are listed in Table 1. ACs support was analyzed most comprehensively since it was the least treated and therefore provided some indication about the metal content of the biomass-based lignin. According to the results, ACs contained minor amounts of metals, such as Ca (0.47 wt%) and Na (0.57 wt%) but its Fe content was very low (0.06 wt%) (Table S1). ACs-based iron-impregnated catalysts 5Fe-ACs and 10Fe-ACs contained 4.0 and 9.2 wt% iron, respectively. Based on these values, iron impregnation was considered successful, as the target amounts were 5 and 10 wt%. The other support, ACz, contained a significant amount of zinc (8.2 wt%), which originated from the chemical activation step in the preparation process. Zinc was also naturally present in ACz-supported 5Fe-ACz and 5FeNO3-ACz (3.8 and 4.6 wt%, respectively) but the amount decreased during the impregnation step. The iron contents for 5Fe-ACz and 5FeNO3-ACz were 4.5 and 5.0 wt%, respectively. Because of the remaining zinc, a third catalyst support ACz N was prepared similarly to ACz but afterwards it was treated with HNO3 in order to remove remaining zinc. According to the ICP-OES results, this treatment did remove zinc because the remaining amount was only 0.07 wt%. Meanwhile, the iron content of 5Fe-ACz N was 5.5 wt%.

Additionally, the surface area (SA) according to Brunauer–Emmett–Teller (BET) theory, average pore diameter and pore volume distributions according to density functional theory (DFT) were determined for all supports and catalysts using N2-physisorption analysis (Table 2). As can be seen from the results, the surface area and pore volume of the chemically activated ACz and ACz N supports were higher (1470/1091 m2g−<sup>1</sup> and 0.72/0.49 cm3g−1) than for the ACs support prepared with steam activation (760 m2g−<sup>1</sup> and 0.47 cm3g−1) (entries 1, 4 and 7). Further, the total pore volume and the relative amount of micropores were higher in ACz and ACz N than in ACs. Treatment with HNO3 after chemical activation decreased the support surface area, average pore diameter and total pore volume but the quantities were still greater than with steam activated ACs.


**Table 2.** N2-physisorption analysis from di fferent activated carbon (AC) supports and catalysts.

\* also contained macro-pores 0.01 cm<sup>3</sup>/g (3%).

The surface area and total pore volume of all catalysts decreased while the iron was impregnated on the carbon surface. This is reasonable, as metal is deposited on the surface and goes into the pores (Table 2). For ACs-based catalysts, the surface area decreased from 860 to 455 or 380 m<sup>2</sup>/g and total pore volume from 0.47 to 0.34 or 0.26 cm<sup>3</sup>/g, depending on how much metal was impregnated on the catalyst surface (entries 1–3). For ACz-based catalysts, the surface area decreased from 1470 to 1000 or 948 m<sup>2</sup>/g and the total pore volume from 0.72 to 0.48 or 0.45 cm<sup>3</sup>/g, depending on which iron precursor was used (entries 4–6). For ACz N-based catalysts, the surface area decreased from 1091 to 790 m<sup>2</sup>/g and the total pore volume decreased from 0.49 to 0.35 (entries 7–8).

Absolute volumes of the meso- and micropores decreased with all catalysts when iron was impregnated but there were di fferences in the final pore volume distributions between meso- and micropores, partly because of di fferent initial pore volume distributions of the di fferent supports. For steam-activated ACs, the micropore volume decreased in 5Fe-ACs and 10Fe-ACs catalysts by 48%–52% compared to ACs. With mesopores, the decrease was smaller, only 12%–38%, meaning that the micropores were primarily filled in ACs when iron was impregnated. Because micropores were more filled with iron than mesopores, the average pore diameter also increased with ACs-based catalysts when compared to plain ACs. Comparing 5Fe-ACs and 10Fe-ACs, the di fference in mesopore volume was notable, which meant that 5% iron addition did not a ffect the mesopores significantly while 10% iron addition filled the mesopores considerably. For chemically activated ACz and ACz N, the micropore volume decreased only 29% or 24%, respectively, while the volume of the mesopores was decreased 39%–52% for ACz and 38% for ACz N. Di fferent precursors did not have a significant effect on surface area or pore volumes. Because the volume of the micropores decreased less than the volume of the mesopores, the relative amount of micropores was increased in chemically activated iron catalysts. In addition, the average pore diameter decreased compared to the supports.

Other surface properties, such as functional groups, acidity and detailed metal composition, were studied using X-ray photoelectron spectroscopy (XPS), the Boehm titration method and X-ray di ffraction (XRD). These analyses were only conducted for ACs, 5Fe-ACs and 10Fe-ACs, which were determined to be the most promising ones in the conversion of xylose to furfural (see Section 2.3). XPS results show that plain ACs support already contains some oxygen functionalities (hydroxyl and carboxyl groups) on the surface but iron addition increases the oxygen–carbon ratio of the catalysts compared to ACs support (Table 3, Table S2). The increase is also dependent on how much iron is added, as the percentage of carbon atoms decreases from 96.8 to 93.6 or 86.3% and that of oxygen atoms increases from 2.9 to 4.1 or 9.5% when the amount of iron increased from 0.06 to 4.0 or to 9.2 wt%, respectively. In addition to C, O and Fe, Cl was detected from iron-impregnated catalysts. An increase of oxygen functionalities was detected from an O1s scan at 531.0eV and at 532.5eV (Table S2). The former can originate from carbonyl groups or metal oxides and the latter, for example, from hydroxyl groups or O-Fe bonds [37,38]. From the XPS Fe2p spectra, a peak at 711.3 eV was detected from both 5Fe-ACs and 10Fe-ACs catalysts, indicating the presence of oxidized iron, Fe2O3 or FeOOH [37,39]. According to XRD, the iron was mostly present as oxides (Fe3O4 and Fe2O3, Figure S1), so it is proposed that oxygen content was increased together with iron content as iron oxide. According to XRD, 10Fe-ACs contained mostly Fe2O3 (hematite, 01-080-5405) and only small amounts of Fe3O4 (magnetite, 04-015-9120). Conversely, 5Fe-ACs contained clearly more Fe3O4 than Fe2O3. However, no iron chlorides were detected with XRD measurements.


**Table 3.** Surface analysis of prepared carbon support and catalysts according to X-ray photoelectron spectroscopy (XPS) and Boehm titration.

> a as atom-%, b by Boehm titration, nd = not detected.

According to Boehm titration, plain ACs contained a small amount of acidic oxygen functionalities (0.07 mmol/g, Table 3, Figure S2.). The amount of acidic oxygen groups increased when the metal was added to 1.77 or 1.95, depending on the metal amount. Acidic oxygen functionalities are probably formed during iron impregnation as a consequence of HCl formation from FeCl3 hydration in water solution. Barroso-Bogeat et al. showed that metal ions markedly influence the pH of the impregnation solution and thereby the oxidizing power of this solution toward the activated carbon support [40]. Since XPS revealed the potential for the presence of iron hydroxides, it is possible that Brønsted acidity of 5Fe-ACs and 10Fe-ACs is induced by iron hydroxides (e.g., FeOOH) [25,41]. Regardless of the specific nature of the acidic oxygen groups, they favor metal adsorption [42].

The morphology of the physically activated support (ACs) and catalysts (5Fe-ACs and 10Fe-ACs) was observed using a scanning electron microscope (SEM) and a scanning transmission electron microscope (STEM). SEM images clearly revealed particles on the carbon surface for 5Fe-ACs and 10Fe-ACs, while plain ACs did not contain any visible particles (Figure 3). The support and both catalysts showed a very porous structure in SEM as well as in STEM. Figure 4 (and Figure S3) shows chemical mapping of the elements C, O, Fe and Cl using energy-dispersive x-ray spectroscopy in scanning transmission electron microscopy (STEM-EDS). Comparing the distribution of Fe and O, it is clear that both elements appear at the same location, which indicates the presence of iron oxide. Therefore, the results obtained from XPS and XRD showing that the iron particles were oxides were confirmed. Mapping also showed that significant amount of residual chlorine was evenly distributed on the surface. Notable chlorine remains have been also detected in the literature when FeCl3 has been used as a metal precursor [43].

**Figure 3.** Scanning electron microscopy (SEM) images of 5Fe-ACs (**a**), 10Fe-ACs (**b**) and ACs (**c**). Iron particles are clearly visible in 5Fe-ACs and 10Fe-ACs while ACs shows a porous structure.

**Figure 4.** Chemical mapping of 5Fe-ACs using energy-dispersive x-ray spectroscopy in scanning transmission electron microscopy (STEM-EDS). The figure reveals that iron is most likely present as iron oxide on the carbon surface, as iron and oxygen appear in the same locations.

#### *2.3. Furfural Production Using Heterogeneous Catalysts*

Conversion studies were started with the control experiment without any support or catalyst (Table 4). This so-called autocatalysis was able to produce a 12% furfural yield and 18% xylose conversion at 160 ◦C in 1.5 h. Autocatalysis is based on high temperature and pressure, where the dissociation constant of water is increased [44]. In addition, formation of organic acids (e.g., formic and lactic acid) during the reaction might occur, which then further catalyzes the hydrolysis reaction [45]. However, a 160 ◦C reaction temperature and 1.5-h reaction time represent rather mild conditions and therefore only low conversion and yield were achieved. All supports and catalysts were able to produce higher furfural yields and conversions than the control experiment. First, chemically activated (ZnCl2) carbon support (ACz) was tested and it resulted in good conversion (82%, Table 4, entry 2). The furfural yield was also considerably high (28%). The high conversion is most likely connected to the high zinc content of ACz (Table 1), which originated from chemical activation. Iron was impregnated to the support using an FeNO3 precursor and FeCl3 precursor (Table 4, entries 3¨C4, respectively). NO3-based salts are commonly favored as precursors since they leave no residue on the catalyst [46,47]. With a nitrate precursor, conversion increased further (from 82 to 91%) compared to ACz support but the yield decreased from 28 to 23% (Table 4, entries 2–3). Therefore, the reaction selectivity toward furfural also decreased (from 36 to 27%). With a chloride precursor, the yield increased (from 28 to 32%) compared to the ACz support and the reaction selectivity also increased (from 36 to 51%, Table 4, entries 2 and 4). Based on the higher furfural yield and reaction selectivity, the FeCl3 precursor was determined to be more suitable than FeNO3 and was used in further catalysts. Similar results were obtained by Chareonlimkun et al., who discovered that chloride-based precursors resulted in higher reactivity compared to nitrate-based precursors in ZrO and TiO catalysts [14]. As mentioned in Section 2.1.2, chloride ions have been shown to enhance furfural formation from xylose by favoring the 1,2-enediol formation before dehydration [34]. This is most probably reason why chlorine-based precursors increase furfural yield and reaction selectivity compared to nitrate-based precursors.


**Table 4.** Furfural yield, xylose conversion and reaction selectivity using various chemically activated catalysts at 160 ◦C with a 1.5-h reaction time.

Based on the results so far (Table 4), the 5Fe-ACz catalyst appeared to be the most promising catalyst. Therefore, it was used in conversion at di fferent temperatures using various reaction times (Table 5, graphical presentation Figure S4). Temperature and reaction time are highly dependent on each other, so the time was increased at each temperature until the conversion reached 98%. At 160 ◦C, a 5 h reaction time was needed to achieve full xylose conversion (98%). In these conditions, the furfural yield was 48%. However, with a shorter 3-h reaction time, a similar furfural yield of 47% was achieved with lower conversion (89%). At a higher reaction temperature of 170 ◦C, the same 98% conversion was achieved in 2.5 h. However, the furfural yield was lower at this temperature with a 2.5 h reaction time (44%) compared to the yield at 160 ◦C but the highest 50% furfural yield was achieved with a shorter 2 h reaction time at 170 ◦C. At 180 ◦C, 99% conversion was obtained in 1.5 h but the furfural yield was low (39%). The highest reaction selectivities (55–56%) were achieved at 160 ◦C with 4 and 3 h reaction times and at 170 ◦C with a 2 h reaction time. At 180 ◦C, reaction selectivity was lowest because of the increasing amount of side reactions. Side reaction products were visible in chromatograms produced by high performance liquid chromatography (HPLC) (Figure S5).


**Table 5.** Effect of reaction temperature and time on furfural yield, xylose conversion and reaction selectivity using 5Fe-ACz as the catalyst. Y=furfural yield, C =xylose conversion, S=reaction selectivity.

Even though good furfural yields and reaction selectivities were achieved with 5Fe-ACz catalyst, there was some problems related to its stability. That is, all zinc already leached out of the catalyst at 160 ◦C in 3 h. In addition, a significant amount of iron (68–78% from initial, depending on reaction conditions) leached out of the catalyst. Therefore, more catalyst supports and catalysts were prepared and tested (Table 6). ACz N was chemically activated similarly to ACz but it was treated with HNO3 after activation in order to remove remaining zinc. It produced a 14% furfural yield and 21% xylose conversion in 1.5 h at 160 ◦C (Table 6, entry 1), which is significantly less compared to ACz. However, the support was now zinc free (Table 1). When iron was impregnated on ACz N, the furfural yield increased from 14 to 22% and the conversion increased from 21 to 47% (Table 6, entries 1–2). The selectivity of 5Fe-ACz N was similar to that of 5Fe-ACz but the furfural yield and conversion were lower (Table 6, entry 2 and Table 4, entry 4). The third AC support, ACs, was physically steam activated with H2O instead of chemical activation. Its surface area was lower than that of ACz or ACz N because of the activation method but the conversion and furfural yields were comparable to ACz N (Table 6, entries 1 and 3). After iron impregnation to ACs, the furfural yield increased from 14 to 25% and the

**Table 6.** Furfural yield, xylose conversion and reaction selectivity using various catalysts at 160 ◦C

conversion increased from 19 to 36%. The reaction selectivity with 5Fe-ACs was significantly higher than with any other iron catalyst: 72% compared to 27/50/51% (Tables 4 and 6).

with a 1.5-h reaction time. **Entry Catalyst Yield (%) Conversion (%) Selectivity (%)**


To summarize the experiments so far, the chemical activation method with the AC support produced the best furfural yields before and after iron impregnation (Tables 4 and 6). However, catalyst characterization revealed that zinc chloride activation left some zinc remains in the catalyst support, which further a ffected its catalytic activity. Furthermore, all the zinc leached out of the catalyst during the first use and therefore the catalyst was suitable for a single use only. HNO3-treated chemically activated catalyst did not contain zinc remains but its catalytic activity in furfural production was comparable to that of physically activated ACs and its selectivity was lower than that of ACs. In addition, the chemical activation of ACz N demands significant amounts of ZnCl2, which is toxic to the environment. Therefore, the most reasonable carbon support for iron impregnation would be ACs, which does not require any chemicals other than water for preparation. As a precursor, FeCl3 was more selective than FeNO3. The reaction temperature of 170 ◦C led to the highest furfural yield and thus it was chosen for further experiments.

Conversion studies were continued with ACs-based catalysts—5Fe-ACs and 10Fe-ACs. In addition, control experiments with plain ACs, without any catalyst and with a similar amount of homogeneous iron (as in 5Fe-ACs) were carried out. Reactions without any catalyst were able to produce at most a 37% furfural yield at 170 ◦C in 3.5 h (Figure 5, black squares). Conversion increased with time and reached 88% at highest. Conversion was notably higher at 170 ◦C than at the lower 160 ◦C reaction temperature and with a shorter 1.5-h reaction time (Table 4). The pH of the water phase also clearly changed to acidic during the reaction. In autocatalyzed reactions, furfural yields are strongly dependent on temperature and time but the highest furfural yields have been found around 50% [6]. Our 37% yield at 170 ◦C in 3.5 h is in good agreemen<sup>t</sup> with the study of Ershova et al., who achieved a 42% furfural yield at 180 ◦C in 3.75 h [10].

A plain activated carbon support (ACs) produced a similar furfural yield as water with all reaction times (Figure 5, red circles). Conversion was slightly higher, probably due to oxygen functionalities on the carbon surface (Table 3), until after3hreaction time it decreased slightly. Additionally, the selectivity was higher with ACs than with plain water. A 40% furfural yield and 77% conversion were the highest achieved with ACs in 3.5 h. When the activated carbon support was impregnated with iron, the furfural yield clearly increased. As expected, iron promoted furfural formation. Yields with both iron catalysts (5Fe-ACs and 10Fe-ACs) were similar with time (Figure 5, blue triangles up and green triangles down, respectively). The yields increased until a 3 h reaction time, after which they leveled o ff to 55–57% depending on the catalyst. The only exception was the 2.5 h reaction with 10Fe-ACs, in which the furfural yield did not increase when compared to the2hreaction. Conversion was similar with both the 5Fe-ACs and 10Fe-ACs catalysts and it increased with time from 59 to 96%. Conversion was clearly higher than with ACs or without a catalyst, which indicated that iron impregnation increased the catalysts' activities. Because the furfural yields and xylose conversions with 10Fe-ACs and 5Fe-ACs were so similar, it was concluded that 4 wt% iron was already enough to increase activated carbon catalyst activity and no benefit was obtained with a higher metal content. In fact, the reaction selectivity was lower with 10Fe-ACs than with 5Fe-ACs. The best reaction selectivity (67%) was achieved with the 5Fe-ACs catalyst in 2 h but it did not decrease notably when the reaction time was increased to 3 h, which resulted in the highest furfural yield.

**Figure 5.** Furfural yield, xylose conversion and reaction selectivity using various catalysts at 170 ◦C.

Based on the previous results, 5Fe-ACs was the most promising catalyst studied. Therefore, it was compared to homogeneous FeCl3·6H2O in similar conditions and using similar amounts of catalytic iron (Figure 5, brown rhombuses). A total of 0.0036 mmol of homogeneous FeCl3·6H2O was able to produce a 27–58% furfural yield in 1–3.5 h. The yield was lower than with 5Fe-ACs until the reaction time reached 3.5 h. Even then, the increase in furfural yield was marginal compared to that with 5Fe-ACs, only one percentage unit. Conversion increased with time from 44 to 85% but was always clearly lower than with 5Fe-ACs. Reaction selectivity also increased with time from 28 to 61% but again it was always lower than with 5Fe-ACs. Better results with carbon-supported 5Fe-ACs than with homogeneous FeCl3·6H2O demonstrated that activated carbon support is a promising option for furfural production. Based on catalyst characterization, iron was oxidized in a heterogeneous catalyst, which may have affected its catalytic activity positively compared to FeCl3. Moreover, hydroxyl groups were detected on the surface of 5Fe-ACs, which increases the catalyst's Brønsted acid sites and therefore can increase furfural production.

The results with 5Fe-ACs (57% yield at 170 ◦C in 3 h) are comparable with studies using carbon-supported titanium catalysts—Mazzotta et al. reported a 51% furfural yield (180 ◦C, 30 min) in a biphasic MeTHF/water system, while Russo et al. achieved a 69% furfural yield (170 ◦C, 3.5 h) in toluene/water [22,23]. The results were higher than with sulfonated SBA- and MCM-supported metal oxide catalysts (SBA-15/ZrO2-Al2O3/SO4<sup>2</sup> and MCM-41/ZrO2/SO42), which resulted in 53 and 50% furfural yields, respectively, at 160 ◦C with a 4-h reaction time [15,16].
